Split (1)

train · 5.09k rows

pmcid stringlengths 6 6	title stringlengths 9 374	abstract stringlengths 2 4.62k ⌀	fulltext stringlengths 167 106k	file_path stringlengths 64 64
516777	Survival of patients with metastatic breast cancer: twenty-year data from two SEER registries	Background Many researchers are interested to know if there are any improvements in recent treatment results for metastatic breast cancer in the community, especially for 10- or 15-year survival. Methods Between 1981 and 1985, 782 and 580 female patients with metastatic breast cancer were extracted respectively from the Connecticut and San Francisco-Oakland registries of the Surveillance, Epidemiology, and End Results (SEER) database. The lognormal statistical method to estimate survival was retrospectively validated since the 15-year cause-specific survival rates could be calculated using the standard life-table actuarial method. Estimated rates were compared to the actuarial data available in 2000. Between 1991 and 1995, further 752 and 632 female patients with metastatic breast cancer were extracted respectively from the Connecticut and San Francisco-Oakland registries. The data were analyzed to estimate the 15-year cause-specific survival rates before the year 2005. Results The 5-year period (1981–1985) was chosen, and patients were followed as a cohort for an additional 3 years. The estimated 15-year cause-specific survival rates were 7.1% (95% confidence interval, CI, 1.8–12.4) and 9.1% (95% CI, 3.8–14.4) by the lognormal model for the two registries of Connecticut and San Francisco-Oakland respectively. Since the SEER database provides follow-up information to the end of the year 2000, actuarial calculation can be performed to confirm (validate) the estimation. The Kaplan-Meier calculation for the 15-year cause-specific survival rates were 8.3% (95% CI, 5.8–10.8) and 7.0% (95% CI, 4.3–9.7) respectively. Using the 1991–1995 5-year period cohort and followed for an additional 3 years, the 15-year cause-specific survival rates were estimated to be 9.1% (95% CI, 3.8–14.4) and 14.7% (95% CI, 9.8–19.6) for the two registries of Connecticut and San Francisco-Oakland respectively. Conclusions For the period 1981–1985, the 15-year cause-specific survival for the Connecticut and the San Francisco-Oakland registries were comparable. For the period 1991–1995, there was not much change in survival for the Connecticut registry patients, but there was an improvement in survival for the San Francisco-Oakland registry patients.	Background Prospective trials have the disadvantages of requiring a long time to complete, and using highly selected patient subgroups in tertiary centers. While one waits for the results to mature, this delays additional research to improve treatment. If there were a method that allowed earlier prediction of the results of prospective trials, advances in cancer treatment could be attained within a shorter time period. There is a parametric lognormal model, proposed by Boag [ 1 - 3 ] that had been retrospectively validated in the literature, and could be used prospectively for clinical trials to predict long-term survival rates several years earlier than would otherwise be possible using the standard life-table/actuarial Kaplan-Meier method of calculation [ 4 ]. The prognosis for metastatic breast cancer is generally poor and therefore it is believed that statistical prediction models for long-term survival rates are not needed. Nevertheless, specific subgroups of metastatic breast cancer patients exist, for which depending on the treatment given, the prognosis is improved so that some patients can survive for some time, particularly for those with limited organs involvement such as involvement with bone and/or skin only. In this situation, for which the present study was relevant, a prediction model, even for metastatic breast cancer, can be useful. Breast cancer, among other cancers, has the highest incidence in women, and many studies are currently in progress to assess treatment regimens. If, even for a subgroup of patients, the 10- and 15-year survival rates can be predicted from follow-up data available only 3 years after a 5-year diagnosis period, this would be a useful means of obtaining study results earlier than would otherwise have been possible. For example, a 15-year survival rate calculated by the Kaplan-Meier method requires at least some patients to have been followed for 15 years. In addition prediction model such as the lognormal model can also be used to review the progress of treatment results for a specific period from a treatment center, and to compare that with another specific period of the same treatment center to evaluate the potential impact for any possible change in treatment policy or guideline. Boag's lognormal model for long-term cancer survival rates has been available for use for some 50 years. When the lognormal model was first proposed in the 1940s, it was difficult to implement because of a lack of computing power, and lack of good quality long-term follow-up data from cancer registries. Since 1970s the model has been used by some authors in breast cancer, cervix uteri cancer, head and neck cancer, intraocular melanoma, choroidal-ciliary body melanoma, and small cell lung cancer [ 5 - 10 ]. Currently, although the computing power is sufficient, good quality follow-up data on a sufficient number of patients are seldom available, and it can be a limitation for its application. Large data registry such as the Surveillance, Epidemiology, and End Results (SEER) data [ 11 ] with good long-term follow-up data available can overcome this potential limitation. Methods Between 1981 and 1985, 782 and 580 female patients of metastatic breast cancer were extracted respectively from the Connecticut and San Francisco-Oakland registries from the SEER database using SEERStat 5.0 software. The two registries were chosen because they are two of the earliest registries, with a large population. The data used were survival time, vital status, cause of death, age at diagnosis, and race. The cause-specific survival was defined as the interval from the date of diagnosis to the date of death from breast cancer or the last follow-up date for censoring purposes, if the patient was alive and still being followed at the time of analysis. The survival time of the uncured group of patients who died of breast cancer had been verified to follow a lognormal distribution previously [ 12 ]. Next, between 1991 and 1995, 752 and 632 female patients of metastatic breast cancer were extracted respectively from the two registries. The data were used to estimate the 15-year cause-specific survival rates before the year 2005. To be comparable, for both the 1981–1985 and 1991–1995 cohorts, the staging system used was the SEER historical system (classified as localized, regional, or distant, based on combined pathologic and clinical data). The choice of 1981–1985 and 1991–1995 has the advantage that the two time periods are not too far apart otherwise there would be too much changes of medical practice. These time periods have a minimum of 5 years follow-up. The overall survival rates (OSR) of the two time periods were calculated using the Kaplan-Meier method. The actual relative survival rates (RSR) were calculated using SEERStat 5.0 software. The modified version of period analysis [ 13 ] was applied using the Hakulinen method [ 14 ] to obtain more up-to-date absolute survival rates (ASR) and relative survival rates (RSR) for comparison purpose with a computer program run by Microsoft Excel software. Validation of the lognormal model The validation of the lognormal model has two phases. Phase 1 tests the goodness of fit to a lognormal distribution of the survival time of those cancer patients who died with their disease present, termed an uncured group with a fraction of 1-C, where C is the cured proportion of patients. The lognormal distribution is similar to the normal distribution in that if the variable in the normal is time t, the variable in the lognormal is the logarithm of t. In other words, the investigators attempt to show that the logarithm of the survival time follows a normal distribution. Phase 2 attempts to show that the lognormal model, using short-term follow-up data, can predict long-term survival rates comparable to those calculated by the Kaplan-Meier life-table method with long-term available. This model can be used to estimate long-term cause-specific survival rates (CSSR) by a maximum likelihood method (e.g., 10-year and 15-year survival rates) from only short-term follow-up data. The maximum likelihood method is used to estimate the CSSR at time τ, and is calculated as [C+(1-C)·Q]·100%, where Q is the integral of the lognormal distribution between the limits time τ and infinity. The lognormal statistical model had been validated in stages III and IV breast cancer in a previous publication that survival rates could be estimated several years earlier than is possible using the standard life-table actuarial method [ 12 ]. The survival time of unsuccessfully treated cases could be represented by a lognormal distribution, the long-term survival rates were predicted by Boag's method using a computer program run by Microsoft Excel. In this parametric lognormal model, the standard deviation S was fixed, and only the two remaining parameters, mean M and proportion cured C, were kept floating when using the maximum likelihood method. Multiple iterations converged to a stable solution for C. A 5-year period of diagnosis could be selected and patients followed as a cohort for an additional 3 years. The current study was for metastatic breast cancer patients treated between 1981 and 1985, with follow-up to the end of year 2000, making the series ideal for validating purposes. For example, for cases diagnosed during the 5-year period, prediction of the 15-year survival rate was made using data at the follow-up cutoff date of December 31, 1988 (i.e., 3 years after 1985). The 15-year survival rate prediction was then validated by Kaplan-Meier life-table calculations using the follow-up data available in 2000. For metastatic breast cancer patients treated between 1991 and 1995, and follow-up to the end of year 2000, prediction of the 15-year survival rate was made using data at the follow-up cutoff date of December 31, 1998 (i.e., 3 years after 1995) before the year 2005. Results From the cohort of 1981–1985 inclusively, 782 patients from the Connecticut registry were followed to the end of 1988. The lognormal model predicted the 15-year CSSR to be 7.1% (95% CI, 1.8–12.4). The 15-year CSSR was 8.3% (95% CI, 5.8–10.8) validated by the Kaplan-Meier calculation using actuarial follow-up data up to the end of year 2000. From the cohort of 1981–1985 inclusively, 580 patients from the San Francisco-Oakland registry were followed to the end of 1988. The lognormal model predicted the 15-year CSSR to be 9.2% (95% CI, 3.9–14.5). The 15-year CSSR was 7.0% (95% CI, 4.3–9.7) validated by the Kaplan-Meier calculation using actuarial follow-up data up to the end of year 2000. Using the same method, the cohort of 1991–1995 inclusively, 752 patients from the Connecticut registry were followed to the end of 1998. The lognormal model predicted the 10-year CSSR to be 12.6% (95% CI, 7.3–17.9). The 10-year CSSR was 11.3% (95% CI, 7.8–14.8) validated by the Kaplan-Meier calculation using actuarial follow-up data up to the end of year 2000. The lognormal model predicted the 15-year CSSR to be 9.1% (95% CI, 3.8–14.4), which cannot be validated before 2005. For the cohort of 1991–1995 inclusively, 632 patients from the San Francisco-Oakland registry were followed to the end of 1998. The lognormal model predicted the 10-year CSSR to be 17.0% (95% CI, 12.1–21.9). The 10-year CSSR was 15.9% (95% CI, 11.4–20.4) validated by the Kaplan-Meier calculation using actuarial follow-up data up to the end of year 2000. The lognormal model predicted the 15-year CSSR to be 14.7% (95% CI, 9.8–19.6), which cannot be validated before 2005. For the period 1991–1995, there was not much change of only about 2% absolute percentage point in the predicted 15-year CSSR for the Connecticut registry, but there was an improvement of about 6% absolute percentage points for the San Francisco-Oakland registry when compared with the period 1981–1985 15-year CSSR, which was validated by the Kaplan-Meier calculation. (Table 1 ) For comparison purpose, the actual OSR and RSR were compared with the ASR and RSR obtained by the period analysis. (Tables 2 and 3 ) It was found that there were more patient survival improvements shown in the actual OSR and RSR for the San Francisco-Oakland registry, but not much for the Connecticut registry. However the period analysis results did not show such improvements. Discussion Lognormal model Rutqvist studied the fit of Boag's lognormal model to the survival times of 8170 breast cancer cases reported to the Swedish Cancer Registry during 1961–1963. The model fitted the 1961–1963 data well for the entire case material and for patients aged less than 70 years. In this registry, the lognormal model did not fit the data for patients aged greater than 70 years, who were more likely to be censored because of coincidental causes of death. Another disadvantage stated by the author was that large number of patients was required to obtain estimates with reasonably small standard errors for breast cancer. With another series of the Norwegian Cancer Registry of 14,000 breast cancer cases, Rutqvist et al . [ 15 ] deduced that lognormal is the best model because other models did not fit the observed survival in all stages, ages, and time periods (two-parameter models, such as exponential or extrapolated actuarial, or three-parameter models, such as sum of two exponential, exponential with shoulder, Weibull). Both the exponential and extrapolated actuarial models assume that the conditional relative survival is lowest immediately after treatment. With the lognormal model, the survival curve has a low initial mortality that rapidly increases to a maximum, with a slow decrease in the mortality after the maximum has occurred. Requirements for using the lognormal model The lognormal model can only predict cause-specific survival, because other coincidental causes of death are too unpredictable (e.g., the rate of stroke). Therefore, overall survival cannot be predicted. The maximum likelihood method is the most accurate method for fitting the lognormal model with the smallest mean squared error. However, there are some requirements for its use. The maximum likelihood method fails to converge to a stable solution using the initial estimates if there is extensive censoring within the data. This occurs if patients are lost to follow-up or die from coincidental non-cancerous causes. The frequency of failure to yield a successful fit for lognormality was greater when one-fourth of cases were designated as lost to follow-up. Gamel et al . established a stable linear algorithm for fitting the lognormal model to survival data. To achieve convergence, some authors have fixed one or two parameters of the lognormal model to pre-selected values to simplify the iterative procedure required for convergence [ 6 ]. Some prognostic factors follow lognormal distribution Prognostic factors in patients with distant metastases at the time of diagnosis were investigated by Rudan et al . [ 16 ], and Chapman et al . [ 17 ], primary tumor size was a significant prognostic factor. Engel et al . [ 18 ] found that the number of metastatic cases and the time to metastasis depended on the tumor diameter at diagnosis. Cell growth is essential for the development of tumors. Tumor size is therefore the most important factor in describing tumor biology. As the tumor size increases, the probability of node-positivity increases. Another study group also found this correlation up to 5 cm [ 19 ]. Tubiana and Koscielny [ 20 ] have found a highly significant correlation between tumor size and the probability of distant metastasis. The distribution of tumor sizes at metastatic spread was lognormal with a median diameter equal to 3.5 cm. The patients were subdivided into 3 groups according to the histological grade. In each subgroup there was a significant correlation between tumor size and the probability of distant spread. The distributions were lognormal and the median size was markedly larger for grade 1 tumors. A number of quantitative postmortem observations regarding the size distribution of metastases have been published [ 21 - 23 ]. These studies revealed a skewed distribution with a high proportion of smaller metastases, and a significant tail extending to the larger metastases, consistent with a lognormal distribution. The more detailed measurements from human liver metastases provided by Yamanami et al . were found to approximate the lognormal distribution reasonably well. A hypothesis was proposed by Kendal [ 24 ] that the time available for the growth of metastases is normally distributed, presumably as a consequence of the summation of multiple independently distributed time intervals from each of the steps and of the Central Limit Theorem. For exponentially growing metastases, the corresponding size distribution would be lognormal; Gompertzian growth would imply a modified (Gompertz-normal) distribution, where larger metastases would occur less frequently as a consequence of a decreased growth rate. These two size distributions were evaluated against 18 human autopsy cases where precise size measurements had been collected from over 3900 macroscopic hematogenous organ metastases. The lognormal distribution provided an approximate agreement. Its main deficiency was a tendency to over-represent metastases greater than 10 mm diameter. These observations supported the hypothesis of normally distributed growth times, and qualified the utility of the lognormal and Gompertz-normal distributions for the size distribution of metastases. Why is the lognormal model applicable to so many organ sites [ 3 , 6 - 10 , 12 , 25 - 36 ] (Table 4 )? Boag's explanation for the lognormal survival time distribution was that if the patient was not cured by treatment, the length of the remaining survival time would be dependent principally on the growth rate of the tumor remnants. Pearlman [ 37 ] estimated the growth rates of breast cancer that recurred in the scar, assuming that the recurrence started from a single cell. He found that the growth rates were approximately lognormally distributed. Likewise, von Fournier et al . [ 38 ] found that the growth rates of breast cancers followed by serial mammography were lognormally distributed. Variation of survival rates over time In order to determine whether current programs for the management of metastatic breast cancer have led to improved patient survival, Debonis et al . [ 39 ] determined the median survival times for five-year intervals of 849 patients admitted to the City of Hope National Medical Center with metastatic breast cancer from 1955 to 1980. Survival times in each of the clinical subsets remained unchanged during the period of observation, regardless of the therapeutic modalities included in the treatment regimens. The study indicates that changes in palliative therapy for metastatic breast cancer during the 25 years of observation have not influenced overall survival. On the contrary, Dickman et al . [ 40 ] studied the survival of cancer patients in Finland during the years 1955–1994. The 5-year RSR for distant metastases breast cancer had increased from 10% for the period 1955–1964 to 22% for 1985–1994. The tumor registry at Yale-New Haven Hospital, which began recording data in 1920, was utilized by Todd et al . [ 41 ] to examine the ultimate outcome of all breast cancer patients who were initially diagnosed at Yale with metastatic breast cancer. The median survival of these patients increased steadily from 21 months in 1920 to 41 months in the decade from 1970 to 1980. The percentage of women actually surviving 5 years increased from 5% in the 1920s to approximately 25% in the 1960s. Despite the use of combination drug programs in the 1970s, the percentage of these patients remaining alive at 5 years remained near 25%. Firm conclusions cannot be made from a retrospective study spanning 60 years, although the trends depicted the lack of continued improvement indicate that the current therapeutic approach to metastatic breast cancer in that period may not result in dramatic improvement in overall survival. Geographical variation of survival rates Farrow et al . [ 42 ] documented substantial geographical variation in patterns of treatment of cancer and other diseases. Because cancer treatment is not uniform nationwide in the States, survival following the diagnosis of cancer might also be expected to vary geographically. Survival data from the nine population-based registries in the SEER Program were analyzed for cancers of the stomach, colon, rectum, lung, breast, uterus, ovary, prostate, and bladder. The patients included all non-Hispanic white patients diagnosed with cancer of one of the selected sites during 1983–1991. Regional variation in crude five-year survival rates across the nine SEER areas was most marked for cancers of the uterus and prostate. For uterine cancer, for example, five-year survival ranged from 73.2% in Connecticut to 84.0% in Hawaii. Less marked variation was observed for cancers of the colon, rectum, and breast. For cancers of the bladder, ovary, stomach, and lung, survival rates five years after diagnosis were relatively invariant across the SEER areas. Maggard et al . [ 43 ] also found that variations in the breast cancer mortality rates exist between states. A nearly 50% increase is observed between the states with the highest and lowest mortality rates. Adjusted analyses demonstrated that stage at presentation is a more important predictor of mortality variation than treatment differences. Goodwin et al . [ 44 ] examined breast cancer incidence, survival, and mortality in the 66 health service areas covered by the SEER program for women aged 65 and older at diagnosis. They found that there was considerable geographic variation in survival from breast cancer among older women, and this contributed to variation in breast cancer mortality. The elevated mortality in the Northeast is apparent only in older women [ 45 ]. For women aged 65 years and older, breast cancer mortality is 26% higher in New England than in the South, while incidence is only 3% higher. Breast cancer mortality for older women by state correlates poorly with incidence (r = 0.28). The above-mentioned results are consistent with that from the present study: the Connecticut registry has lower CSSR than the San Francisco-Oakland registry for the period 1991–1995. The Connecticut cohort has median age at diagnosis of 66 (range 25–103), while the San Francisco-Oakland cohort has lower median age of 63 (range 26–96). It could be argued that new treatments evolved in the recent decade have improved the survival of the breast cancer patients, and younger patients benefit more than the older patients. Apart from treatment offered, changes of survival rates over time or geographical areas can be due to co-morbidities or other characteristics such as race, age, and differences in staging procedures. Conclusions For the period 1981–1985, the 15-year cause-specific survival for the Connecticut and the San Francisco-Oakland registries were comparable. For the period 1991–1995, there was not much change in survival for the Connecticut registry patients, but there was an improvement in survival for the San Francisco-Oakland registry patients. List of abbreviations CSSR: Cause-specific survival. SEER: Surveillance, Epidemiology, and End Results. OSR: Overall survival rate. ASR: Absolute survival rate. RSR: Relative survival rate. Competing interests None declared. Authors contributions PT: Data analysis and writing of manuscript. EY, VVH, GC, GV: Critical appraisal of manuscript. All authors read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC516777.xml
532400	A Global View of Gene Expression in the Aging Kidney	null	Four years ago in Science , Stuart Kim, a Stanford developmental biologist, made the case for laying down the broad strokes of a complex physiological process before defining its mechanisms. “A powerful, top-down, holistic approach,” he wrote, “is to identify all of the components of a particular cellular process, so that one can define the global picture first and then use it as a framework to understand how the individual components of the process fit together.” To get a broad view of gene expression in the aging nematode, Kim's lab turned to DNA microarrays and functional genomics. In a new study, Kim and colleagues apply this same approach to the decidedly more complex problem of human aging and “present a molecular portrait” of the aging kidney. Scientists have identified a wide range of molecular pathways and mechanisms associated with aging. Many have been found in evolutionarily distant organisms, suggesting they have been conserved and could shed light on human aging. Yet other studies suggest that since few animals reach old age in the wild, any aging-related physiological changes aren't likely to impact the fitness of a population and so aren't likely to be conserved. Consequently, aging pathways in worms, for example, would have little bearing on humans. To investigate the molecular pathways associated with human aging, the authors focused on human tissue—in this case, the kidney. Kidneys came from 74 patients, ranging in age from 27 to 92. Samples were extracted from donated kidneys or “meticulously harvested” from kidneys with localized disease to ensure only normal tissue was taken. Two structures that are critical to kidney function (removing toxins from blood) were removed from each sample: the renal cortex, which filters plasma, and the medulla, which alters urine composition to maintain fluid balance. Both deteriorate with age. An extensive clinical history was noted for each sample to account for any potentially confounding medical factors. Transcriptional profiling to study aging in the kidney Kim and colleagues then isolated RNA transcripts from the samples to determine the activity of every gene, broken down by age and kidney section, through microarray analysis. Looking for differences in gene expression across the genome, they identified genes that showed a statistically significant change in expression as a function of age. Of 33,000 known human genes on the microarray, 985 showed age-related changes, most showing increased activity. These changes are truly age-regulated, the authors conclude, since none of the medical factors impacted the observed changes in gene expression. Although cortex and medulla have different cell types and perform different functions, their genetic aging profile was very similar, suggesting a common aging mechanism operates in both structures. In fact, these mechanisms may function broadly, as most of the age-regulated kidney genes were also active in a wide range of human tissues. Other organisms appear to lack these changes, however, prompting the authors to argue that understanding aging in humans will require human subjects. Most importantly, the genetic profile of the tissue samples correlated with the physiological and morphological decline of an aging kidney. An 81-year-old patient with an unusually healthy kidney had a molecular profile typical of someone much younger, while a 78-year-old with a damaged kidney had the profile of a much older person. Using the power of functional genomics, this study has identified a set of genes that can serve as molecular markers for various stages of a deteriorating kidney and predict the relative health of a patient compared to their age group. These gene sets can also serve as probes to shed light on the molecular pathways at work in the aging kidney, and possibly on the process of aging itself.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC532400.xml
509287	Cancer immunotherapy: avoiding the road to perdition	The hypothesis that human cancers express antigens that can be specifically targeted by cell mediated immunity has become a scientifically justifiable rationale for the design and clinical testing of novel tumor-associated antigens (TAA). Although a number of TAA have been recognized and it has been suggested that they could be useful in the immunological treatment of cancer, the complexity of human beings leads us to reflect on the need to establish new criteria for validating their real applicability. Herein, we show a system level-based approach that includes morphological and molecular techniques, which is specifically required to improve the capacity to produce desired results and to allow cancer immunotherapy to re-emerge from the mist in which it is currently shrouded.	Introduction Although considerable advances have been made in terms of our molecular and cellular knowledge, for most human disease states a fundamental understanding of causal disease onset, disease mechanism and progression, and optimal treatment is still significantly limited. In part, this advancement has been hampered by our inability to fully and rapidly delineate complex cellular metabolic processes and molecular pathways. Organisms are complex self-organizing entities made up of such parts: organs, tissues, cells, organelles and ultimately molecules and atoms. One question that arises, concerns the relationship between the whole and its component parts. The issue at stake is sometimes called "the question of reduction" or "the problem of reductionism" [ 1 ]. The inefficacy of contemporary science to describe biological systems, consisting of non-identical parts that have different and non-local interactions has tended to limit progress in the human healthcare. Many biological systems remain incomprehensible because their multifarious nature has been combined with a reductionist approach based on the linear conception of cause and effect . The use, however, of a more holistic multidimensional system level-based approach may provide new insights into the understanding of disease processes and mechanisms of action of therapeutical agents [ 2 ]. Herein we aim to introduce a system level-based approach that includes morphological and molecular techniques for validating the appropriateness of using novel tumor-associated antigens (TAA) for clinical purposes. This approach might be easily implemented for identifying prognostic, diagnostic and alternative biomarkers. Finally, this type of analysis of appropriately designed cohorts might also provide a key to understanding the differences in patients who do or do not respond to any particular therapy. This information may be helpful for a more effective (and therefore more cost-effective ) design of clinical trials [ 2 ]. Immunotherapy and the human complexity The recognition and characterization of novel TAA is fundamental to the advance of cancer immunotherapy. The original hypothesis of Boon [ 3 ] and Rosenberg [ 4 ] that human cancers express antigens that can be specifically targeted by cell mediated immunity has become a scientifically justifiable rationale for the design and clinical testing of novel TAA based immunotherapies and therapeutic vaccines [ 5 - 7 ]. However, although a number of TAA have been discovered and it has been suggested that they could be useful in the immunological treatment of cancer, the complexity of human beings leads us to reflect on the need to establish compelling new criteria for validating their real applicability. Biological complexity can be intuitively appreciated – at least in terms of morphological or behavioral complexity, or the variety of cell types in an organism – but the term itself is notoriously difficult to define [ 8 ]. Human beings are complex hierarchical system s consisting of a number of levels of anatomical organization (genes, cells, tissues, organs, apparatuses, and organism) that interrelate differently with each other to form networks of growing complexity. The concept of anatomical entities as hierarchy of graduated forms, and the increasing number of known structural variables, have highlighted new properties of organized biological matter and raised a series of intriguing questions. In order to understand biology at the system level , we need to examine the structure and dynamics of the functions of organisms rather than the characteristics of their constitutive isolated parts [ 8 - 13 ]. The expression of TAA in biological materials has mainly been studied at the level of gene expression and gene level measurement by Reverse Transcriptase-Polymerase Chain Reaction (RT-PCR) analysis and the Quantitative real-time PCR (qrt-PCR) technology [ 14 - 17 ]. However, the information provided by these approaches is limited by the fact that the phenomena observed at each level of anatomical organization have properties that do not exist at a lower or higher level: RT-PCR and qrt-PCR may offer a satisfactory qualitative/quantitative description of small-scale structures , but this is likely to be irrelevant when it comes to large-scale features . The above considerations, in conjunction with the complexity of tumor-host interactions within the tumor microenvironment caused by temporal changes in tumor phenotypes and an array of immune mediators expressed in the tumor microenvironment [ 18 ] might clarify the limited reliability and applicability of current immunotherapeutic approaches. Here, we suggest a system level-based approach (Figure 1 ) for validating the appropriateness of using TAA for clinical purposes, which includes the following never defined before key points: Figure 1 System level-based approach for validating candidate TAA for clinical application. Aside from the well defined experimental procedure, the method presented here is based on the complex hierarchical nature of the human beings. The analysis begins at level of gene expression and then continues to higher levels of anatomical organization, (cell, tissue, organ, apparatuses and organism). This approach includes both morphological and molecular techniques. It also introduces the concept of dynamics of TAA expression at the level of the cell cycle , the physiological status of the organism and the process of aging . • Discriminating the cell types expressing the candidate antigen on the basis of the morphological visualization of all of the parts making up the organ under investigation. • Discriminating the candidate antigen's sub-cellular localization (at the level of cell nucleus , cytoplasm and/or plasma membrane ) by ultra-structural morphological visualizations. • Mapping candidate antigen expression in all of the organs making up the apparatuses . • Mapping candidate antigen expression in all of the apparatuses making up the living organism . • Estimating the percentage of normal cells and their neoplastic counterparts expressing the candidate antigen. • Evaluating the dynamics of candidate antigen expression at the level of the cell cycle , the physiological status of the organism ( i.e. the woman's menstrual cycle) and the process of aging . In order to advance our knowledge in a currently widely debated field of investigation, a clearer distinction must be made between in vitro laboratory results (the discovery and validation of target antigens) and their in vivo application ( in vivo validation), and it is necessary to adopt a more complete experimental approach that forcefully includes both morphological and molecular techniques [ 19 ]. Conclusions Translational science which is aimed to test, in humans, novel therapeutic strategies developed through experimentation [ 20 ] should begin to consider the role of emergence in other words the appearance of unexpected structures and/or the occurrence of surprising behaviors in large systems composed from microscopic parts, whether physical or biological. By unexpected and surprising we mean structures and behaviors which are not intuitive and are not simply predictable . Since our understanding of complex human disease such as cancer, is still limited and pre-clinical models have shown a discouraging propensity [ 2 , 6 ] to fail when applied to humans, a new way of thinking is strongly needed that unites physicians, biologists, mathematicians and epidemiologists, in order to develop a better theoretical framework of tumor development, progression and tumor-host interactions. Although the model presented here is based on a multidisciplinary system-level approach probably within the reach of only very large and multi-talented laboratories, it is aimed to introduce a different way of investigating human cancer, which takes into account the complexity of the human being as a system. The use of a holistic approach, which enables a more accurate selection of immunotherapeutic target antigens in the first phase of the experimental research, will reduce the notable fragmentation of the biological information in the post-genomic era, and will facilitate a more accurate transfer of the acquired knowledge to the bedside. Further, this new multidisciplinary approach is specifically required to improve the capacity to produce desired results with a minimum expenditure of energy , time , or resources for immunotherapeutic treatments and to allow cancer immunotherapy to re-emerge from the mist in which it is currently shrouded.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC509287.xml
535540	Weather-based prediction of Plasmodium falciparum malaria in epidemic-prone regions of Ethiopia I. Patterns of lagged weather effects reflect biological mechanisms	Background Malaria epidemics due to Plasmodium falciparum are reported frequently in the East African highlands with high case fatality rates. There have been formal attempts to predict epidemics by the use of climatic variables that are predictors of transmission potential. However, little consensus has emerged about the relative importance and predictive value of different factors. Understanding the reasons for variation is crucial to determining specific and important indicators for epidemic prediction. The impact of temperature on the duration of a mosquito's life cycle and the sporogonic phase of the parasite could explain the inconsistent findings. Methods Daily average number of cases was modeled using a robust Poisson regression with rainfall, minimum temperature and maximum temperatures as explanatory variables in a polynomial distributed lag model in 10 districts of Ethiopia. To improve reliability and generalizability within similar climatic conditions, we grouped the districts into two climatic zones, hot and cold. Results In cold districts, rainfall was associated with a delayed increase in malaria cases, while the association in the hot districts occurred at relatively shorter lags. In cold districts, minimum temperature was associated with malaria cases with a delayed effect. In hot districts, the effect of minimum temperature was non-significant at most lags, and much of its contribution was relatively immediate. Conclusions The interaction between climatic factors and their biological influence on mosquito and parasite life cycle is a key factor in the association between weather and malaria. These factors should be considered in the development of malaria early warning system.	Background Malaria epidemics due to Plasmodium falciparum are reported frequently in the East African highlands [ 1 - 6 ]. Immunity to malaria in the populations of these epidemic-prone regions is often incomplete, so that epidemics cause high case fatality rates among all age groups. In 1958, a malaria epidemic covering over 250,000 square kilometers resulted in an estimated three million cases and 150,000 deaths in Ethiopia [ 2 ]. Since then, large scale epidemics of malaria have been noted every five to eight years. Thus, there is an urgent need for the development of malaria early warning systems [ 7 - 9 ] to predict where and when malaria epidemics will occur, with adequate lead-time to target scarce resources for prevention activities. Unusual meteorological conditions, such as especially high rainfall or high temperature, are often cited retrospectively as the precipitating factors for epidemics [ 10 , 11 ]. There have also been formal attempts to predict epidemics by the use of local weather and/or global climatic variables that are predictors of vector abundance and, therefore, of transmission potential [ 12 - 20 ]. However, little consensus has emerged about the relative importance and predictive value of different factors [ 2 , 6 , 21 - 27 ]. Woube [ 27 ] showed that although one epidemic in Ethiopia was associated with higher rainfall, an epidemic in another year was preceded by very little rainfall. Lindsay et al. [ 23 ] found a reduction in malaria infection in the Usambara Mountains of Tanzania following 2.4 times more rainfall than normal, while excessive rainfall during the same period was associated with increased malaria in south-western highlands of Uganda [ 3 ]. Moreover, Mbogo et al. [ 24 ] found variation in the relationship between the mosquito population and rainfall in different districts of Kenya and attributed the variation to environmental heterogeneity. Similarly, Zhou et al. [ 28 ] showed that there was high spatial variation in the sensitivity of malaria outpatient numbers to climate fluctuations in East African highlands. Similarly, determination of the amount of lead time between weather factors and malaria cases is necessary to develop prediction models, but results from different studies have revealed a range of lead times [ 3 , 4 , 22 , 24 , 29 ]. Despite the varying results of these studies, there has not been critical examination of the sources of variation in the association and lag structure (the magnitude of association between weather and malaria at a later time) between weather and incidence of malaria. The inconsistency of findings from different studies may be due in part to the interaction of weather factors affecting vector abundance and survival, and parasite maturation, key determinants of malaria transmission. Deposition of mosquito eggs, and their maturation into larvae and then into adults, requires aquatic breeding sites, and is, therefore, dependent on rainfall [ 30 , 31 ]. The time required for mosquito maturation shortens as temperature increases [ 30 , 32 ]. At 16°C, larval development may take more than 45 days (reducing the number of mosquito generations and putting the larvae at increased risk of predators), compared to only 10 days at 30°C (Table 1 ). By affecting the duration of the aquatic stage of the mosquito life cycle, temperature determines the timing and abundance of mosquitoes following adequate rainfall. Table 1 The effect of mean temperature on the duration of mosquito's life cycle and sporogonic cycle and its effect on the amount of lead time from the availability of breeding sites to the occurrence of malaria cases. Weather factors Stages and duration of mosquito's life cycle and sporogony cycle affected by weather factors Mean temperature (Rainfall temperature) Availability of breeding sites ----------------→ Malaria Mosquito's life cycle* Sporogony† Incubation period in human host Larva ----→ Adult(days) Adult first bite ----→ Infectious bite (days) 16°C 47 111 (10–16 days) 17°C 37 56 18°C 31 28 20°C 23 19 22°C 18 7.9 30°C 10 5.8 35°C 7.9 4.8 39°C 6.7 4.8 40°C 6.5 4.8 * see references [32, 44]; †see references [35, 45] Once female adult mosquitoes emerge, they look for a blood meal, and in the process they ingest malaria parasites (gametocytes) with the blood. The feeding frequency of mosquitoes increases with temperature, resulting in increased proportions of infective mosquitoes [ 33 ]. The duration of the extrinsic phase of the parasite (sporogony cycle), which is the development of the ookinete, the egg of the parasite, in the midgut of the anopheline mosquito, depends on temperature. The sporogony cycle on average lasts about 10 days, but shortens as temperature increases [ 34 , 35 ], becoming as short as five days when the temperature exceeds 30°C (Table 1 ). These data on the timing of the mosquito life cycle suggest that malaria cases should follow, at defined intervals, periods of increased temperature and increased rainfall. Moreover, because temperature accelerates several steps in the process of mosquito and parasite development, the time lag between the appearance of suitable weather conditions and the appearance of new malaria cases should shorten as temperature rises. Although based largely on laboratory findings, these data suggest quantitative hypotheses about the precise time lag between increases in temperature and rainfall, and increases in malaria cases. At an average temperature of 20°C, the aquatic phase of the mosquito will be completed in about 28 days (five days for eggs to hatch and 23 days for the larva to develop into adult stage); and sporogony is completed in 28 days (Table 1 ). At this temperature, malaria cases should, therefore, appear 9–10 weeks following rainfall, assuming an average incubation period of about 10–16 days. Similarly, in this situation, the number of malaria cases should be positively related to increases in temperature three to seven weeks beforehand (during the aquatic and sporogony stages). When the mean temperature is higher (30°C), the aquatic stage of mosquito and the sporogony cycle are completed in about 12 and 8 days respectively (Table 1 ). In this relatively hot environment, malaria cases should appear 4–5 weeks following rainfall and the lag in the effect of temperature should also be shorter. The purpose of the investigations reported here is to test these hypotheses using weekly data on weather and malaria cases. Specifically, we used Polynomial Distributed Lag Models (PDL) to determine the effect of weather factors and their lag distribution on malaria in relatively hot and cold environments using a data set consisting of weekly parasitologically confirmed malaria cases collected from health facilities and locally collected meteorological factors in 10 districts of Ethiopia for the years 1990 to 2000. Methods Data Microscopically confirmed malaria cases were collected from a health facility in each of ten districts of Ethiopia over an average of 10 years. Each of these health facilities serve people living in the surrounding localities with few exceptions coming from other places. The data were extracted from records of outpatient consultations for the years 1990 through 2000. These data were compiled by residency (urban and rural) and Plasmodium species; the analysis was restricted to P. falciparum . The original data collected on the basis of Ethiopian weeks were normalized to obtain mean daily cases for each Ethiopian week [ 36 ]. Daily meteorological data (minimum and maximum temperature and rainfall) recorded at the local weather stations nearest to the health facility were obtained from the National Meteorological Services Agency (NMSA) for the same period. The assumption is that the meteorological data from the local weather stations represent the surrounding localities whose inhabitants are served by the respective health facilities. These daily data were collapsed into weekly data to correspond with the weekly malaria cases. The weekly mean for minimum and maximum temperature and the total weekly amount of rainfall were calculated from the daily records. Missing data The data set consists of records with missing values for some of the variables (4.4%, 10.2% and 9.9% for rainfall, minimum and maximum temperature respectively). Because of the multiple time lags considered in the analysis (see below), discarding estimates with a single missing value results in multiple records lost for each missing value. To avoid this substantial loss of data, missing values of the independent variables (weather variables) were interpolated by fitting a linear regression using the value from the previous week and a dummy variable for that week. A new variable was created with original values for non-missing data points and interpolated values for missing data points. Model The daily average number of cases (of the weekly microscopically confirmed malaria cases) was modeled using a robust Poisson regression (implemented in Splus), using weekly rainfall, minimum and maximum temperature as explanatory variables in a distributed lag model. The model was: where E ( Y st ) denotes the expected value for the daily average number of malaria cases at site s on week t ; , and R t - i are the weekly minimum and maximum temperature and the rainfall i weeks previously; α s and β s represent the intercept and coefficient for time trend ( t ), which are specific to the site under consideration. As described in the introduction, biological considerations suggest that the lag between rainfall and associated malaria cases will be different from that between temperature and associated cases. Therefore, lags of 4–12 weeks were considered for rainfall, and relatively shorter lags of 3–10 weeks for the minimum and maximum temperature. The assumption is that changes in temperature and rainfall at a particular time do not have an important influence after 10 and 12 weeks respectively. The overall effect of a unit increase in minimum temperature (for example) is the sum of the coefficients β i . A large number of variables (a total of 25; eight lags each for minimum and maximum temperatures and nine lags for rainfall) were introduced for the three weather factors in this model. The efficiency of the coefficient estimates may be affected due to the large number of parameters to be estimated and the possibility of multicollinearity between the lagged weather factors. When lag terms are put in the same model, correlation between measurements of weather on weeks close together will cause a high degree of collinearity that may result in unstable estimates. Polynomial Distributed Lag Model A polynomial distributed lag (PDL) model [ 37 ] imposes constraints on the coefficients β i , γ i , θ i , forcing each of them to take the form of a (separate) 4 th -degree polynomial in i. This reduces the number of degrees of freedom for each weather factor from the number of lags considered to five and circumvents some of the difficulties associated with estimation of coefficients for multiple lags, including instability of estimates due to collinearity of the different lags of the same variable. With this model the coefficients for lagged minimum temperature, for example, are assumed to take the form: where φ k is the parameters of the d-th degree (here d = 4) polynomial distributed lag. To estimate the parameters describing the polynomial lag φ k equation (2) was substituted into the unconstrained distributed lag model (1) to obtain a constrained polynomial distributed lag model: Grouping of districts The effects of weather factors on the number of malaria cases were distributed over multiple weeks and the separate analysis for each district (not shown) indicated heterogeneity in magnitude and direction of the effects of the weather factors. Districts with similar climatic characteristics were grouped, in order to reduce the effect of random error and to produce more reliable and precise estimates of weather effects. Moreover, this approach will produce more generalizable results within similar climatic conditions. Thus, the districts were grouped into hot (altitude < 1700 mm above sea level) and cold on the basis of altitude and temperature. The hot districts included Diredawa, Nazareth, Wolayita and Zeway; and the cold districts included Alaba, Awasa, Bahirdar, Debrezeit, Hosana and Jimma. Two separate PDL models were fitted to estimate the effects of different weather factors. A basic issue in epidemiological analysis is controlling for the effect of confounding factors. Malaria transmission is affected by different factors and shows a systematic variation over time. To control for other factors that may affect the long-term trend, a time variable was included in the model, which would remove the long-term wavelength patterns, leaving the deviations representing short fluctuations. Since the long-term trend and the numbers of weekly cases vary between districts, an interaction term between the time variable and district (dummy variable) was introduced. Urban areas may have other sources of breeding sites for mosquito not driven by rainfall. To examine the influence of these and other unmeasured factors that vary between urban and rural environments, separate models were fitted for urban and rural cases. Results The data set consists of microscopically confirmed P. falciparum cases from a health facility in each of 10 districts over an average of 10 years and meteorological data from local stations in each of the districts. Table 2 presents the descriptive analysis of cases, meteorological variables and altitude of each district. The daily averages treated by each of the 10 health facilities ranged from 11–39 malaria cases and over 300 cases during the peak transmission season. Minimum temperature was positively correlated with rainfall, significantly (rho = 0.37) in the cold districts and nonsignificantly (rho = 0.06) in the hot. Maximum temperature, however, was negatively correlated with rainfall, significantly (rho = -0.33) in the cold districts and nonsignificantly (rho = -0.033) in the hot. Table 2 Characteristics of the study districts (average daily malaria cases and meteorological variables) District Altitude Malaria cases Maxt Mint Rain Group (m) Mean Min Max (°C) (°C) (mm) Alaba 1750 39 0 163 27.4 11.7 19.6 cold Awasa 1750 11.3 0 17 27.4 12.6 19.3 cold Bahirdar 1770 22.1 1 83 27.2 13 25.2 cold Debrezeit 1900 25.2 1 147 26.5 12.1 16.5 cold Hosana 2200 19.4 0 96 22.4 10.7 23.4 cold Jima 1725 13.2 0 85 27.6 11.6 30 cold Diredawa 1260 25.3 0 330 31.8 19 13.8 hot Nazareth 1622 17.7 0 109 27.5 14.3 17.1 hot Wolayita 13.9 0 113 25.3 14.4 24.4 hot Zeway 1640 11.4 1 102 27.5 13.9 13.7 hot maxt-maximum temperature, mint-minimum temperature The effect of rainfall and temperature on daily average microscopically confirmed cases was estimated by lag in the 10 districts grouped into two climatic zones, hot and cold (Table 2 ). Figure 1a shows the estimates of the distributed lag between rainfall and cases in cold areas. Coefficients represent the multiplicative effect of one additional millimeter of rain at a given lag on the number of malaria cases at a site. Rainfall is significantly associated with the number of malaria cases in the cold districts. The magnitude and direction of the association varies with lags. At shorter lags of 4 and 5 weeks, rainfall is negatively and significantly associated with malaria cases. At lags of six, seven and eight, rainfall is not significantly associated with malaria cases. Lags of nine, 10, 11 and 12 are positively associated with malaria cases and the magnitude of effect increases almost linearly with maximum effect at lag 12. The conclusion is that rainfall in the cold districts is associated with a much delayed increased malaria cases and immediate decrease in malaria cases. Figure 1 Distributed lag structure for the association between 1 mm increase in rainfall, 1°C increase in minimum and maximum temperature and average daily malaria cases. (a) & (b) for rainfall, (c) & (d) for minimum temperature, and (e) & (f) for maximum temperature in the cold and hot districts respectively. The shaded areas represent 95% confidence intervals. Similarly, rainfall is significantly associated with malaria cases in the hot districts. Compared to the cold districts, a significant and positive effect of rainfall in the hot districts manifests at relatively shorter lags (six, seven, eight, nine and ten weeks) and remains positive afterwards but declines and becomes non-significant for the longer lags (Figure 1b ). Much of the contribution of rainfall to the increase in malaria cases in the hot districts occurs at relatively shorter lags (compared to its effect in cold districts) and wanes slowly with increasing lags. Thus, the results for rainfall agree qualitatively with biological expectations. Figure 1c shows the estimated distributed lag relationship between minimum temperature and malaria cases in the cold districts; coefficients represent the multiplicative effect of one degree Celsius increase in temperature at a given lag on the number of malaria cases at a site. Minimum temperature is positively associated with the number of malaria cases, with a significant increase extending from 7 to 10 weeks prior to cases and the size of the effect growing over that range. In the hot districts, by contrast, the effect of minimum temperature on malaria cases is more complicated. At short lags, its effect is small and non-significantly positive (Figure 1d ). A significant positive association at longer lags is also observed. In summary, minimum temperature contributed significantly to the estimated increase in malaria cases in the cold districts with a delayed effect. In the hot districts, while its effect is non-significant, much of the contribution is relatively immediate. Unexpectedly, the only significant contribution of minimum temperature in the hotter districts occurs at long lags. Figures 1e and 1f show the relationship between maximum temperature and malaria cases in the cold and hot district groups respectively. Maximum temperature is not significantly associated with the estimate of malaria cases in either group of districts. However, the trend of the estimates along the lags shows that at shorter lags of three, four and five weeks, maximum temperature is negatively associated with the number of malaria cases in hot districts, while it is positively associated at lags of six, seven and eight weeks in the cold districts. To test the linearity of the association between the weather factors and malaria cases, a three dimensional relationship between weather factors, lag and the magnitude of effect at each lag was explored (See Additional file 1 for the method and figure). In that figure, a positive effect of a factor at a given lag is seen as a positive slope of the surface cut at the given lag; the magnitude of that slope corresponds to the linear effect estimated by the PDL model. The effect of rainfall plateaus at higher rainfall levels; beyond a given quantity of rain, additional rain adds little to the malaria risk. Similarly, although the effect of minimum temperature in the cold districts is linear, it levels off at higher temperature (>16°C) in the hot districts. The results of analysis stratified by rural versus urban sites are shown in Figure 2 . The association between rainfall and cases varies by residency (a & b). Rainfall is significantly associated with cases originated from rural residents but not generally among urban residents, however, the magnitude of effect of rainfall looks similar in both rural and urban areas,. The effect of minimum temperature on malaria cases does not vary by residency (c & d). Figure 2 Distributed lag structure of the effects of rainfall and minimum temperature on average daily malaria cases by residency. (a) & (b) for rainfall in rural and urban respectively, (c) & (d) for minimum temperature in rural and urban respectively. The shaded areas represent 95% confidence intervals. Predicted case numbers for hot and cold districts were compared against actual values to assess how well the models predict the seasonal peaks and interannual variability. Plots of the actual data and predicted values showed that the models predicted the seasonal fluctuations very well (Figure 3 ). However, the models were not able to differentiate clearly between years with high and low peaks. Figure 3 Plot of observed number of cases and predicted cases from the polynomial distributed lag models of 4 districts. Discussion The development of malaria early warning systems [ 7 , 8 ] to predict where and when malaria epidemics will occur, in order to target scarce resources for prevention activities [ 9 , 38 ], has motivated many studies [ 13 - 15 , 18 , 19 ]. However, little consensus has emerged as to which factors should be used as indicators, because multiple studies have yielded differing results on the main determinants of increased malaria transmission [ 2 , 6 , 21 - 23 , 25 , 27 ] and the lead time prior to observable effects [ 3 , 4 , 22 , 24 , 29 ]. In this study a polynomial distributed lag model was used to assess the lag distribution of the effects of weather factors on Plasmodium falciparum malaria in relatively hot (Diredawa, Nazareth, Wolayita and Zeway) and cold (Alaba, Awasa, Bahirdar, Debrezeit, Hosana and Jimma) environments in Ethiopia. The findings are largely consistent with hypotheses based on the relationship between weather factors and mosquito and parasite development. Rainfall is associated with malaria cases in both hot and cold districts with a lagged effect, and as expected, this lag is shorter in hot districts. The effect of rainfall on malaria is linear with saturating effects at higher rainfall levels (See Additional file 1 for the method and figure). Interestingly, malaria in the urban areas is not associated with rainfall. Although the maximum temperature is not generally associated with malaria cases in either group of districts, the minimum temperature is significantly associated with malaria cases in the cold districts with delayed effect, and the lag for the minimum temperature is shorter than that for rainfall, reflecting the two factors' effects on different stages of the transmission cycle. The detection of a positive effect of the minimum temperature at long lags (9–10 weeks) in the warmer districts was not predicted by biological considerations. One of the most striking uncertainties in the literature on weather and malaria is the variability in the reported relationship between rainfall and malaria, with several studies showing the importance of rainfall as a precipitating factor for malaria transmission [ 3 , 4 , 10 , 11 , 29 ], while other studies show negative or neutral effects [ 21 , 23 , 26 , 27 ]. For rainfall to have a positive effect on malaria cases, the temperature must be warm enough to support mosquito and parasite development [ 39 ], and, as the data confirm, the effect of rainfall on cases becomes more immediate in warmer temperatures. This is consistent with the laboratory findings that a mosquito population peaks early at higher temperatures, while a mosquito population at low temperatures experiences slow, steady growth with a delayed peak [ 40 ]. Increases in rainfall may also fail to produce additional malaria cases if aquatic breeding sites are not limiting for mosquitoes; this mechanism is consistent with the observed saturating effect of rainfall in our data. Furthermore, malaria in the urban areas is not significantly associated with rainfall, which may have been one of the sources of inconsistent findings of such analysis. The weak association may be due to the presence of other sources of breeding sites that may persist during the dry season such as brick pits, puddles, blocked drains and cisterns [ 41 ]. Moreover, developmental activities, aggregation of migrant labor forces and overall population movement affect urban malaria. It is also interesting to note that the effect of rainfall in the cold districts is negative at shorter lags, which may be due to breeding sites being flushed away during the rainy season [ 23 ]. Another possible explanation for the negative effect could be that low temperature during the rainy season might suppress malaria transmission. Maximum temperature was lower during the rainy season (shown by negative correlation with rainfall), however, the effect of maximum temperature on malaria is non significant. Moreover, minimum temperature seems to be elevated during the rainy season (positive correlation). Although the effect of rainfall in the hot districts declines after longer lags (due to evaporation and drying up of breeding sites), making the main transmission season shorter, the overall effect of rainfall (sum of the lag coefficients) is bigger in the hot than cold districts. Taken together, the analyses suggest that temperature requirements, saturating effects of rainfall, and urban-rural differences in the effect of rain on malaria transmission are all plausible mechanisms that could explain the inconsistent relationship between excessive rainfall and malaria epidemics. The minimum temperature contributes significantly to the estimated increase in malaria cases with a delayed effect in the cold districts, but not in the hot districts. At lower temperatures, the larval and pupal stages of mosquitoes take longer to complete (for example, 47 days at 16°C) and a small increase in temperature substantially shortens the duration of these phases (to 37 days at 17°C). Similarly, the duration of the sporogony cycle will be short with increasing temperatures (Table 1 ). In addition, raised temperature increases the frequency of mosquito feeding and, hence, the probability of transmitting infection [ 33 ]. Although all such effects of minimum temperature increase malaria transmission in the cold districts, the effect will be seen after a lag. The effect of minimum temperature in the hot districts, on the other hand, is immediate but non-significant. These findings are consistent with reports that small increases in temperature will have a greater effect on malaria transmission in areas with relatively lower average temperatures than areas with higher temperatures [ 42 , 43 ]. The significant effect of minimum temperature at relatively long (9–10 week) lags is not explainable, to our knowledge, on biological grounds. The maximum temperature is not significantly associated with cases in either hot or cold districts. However, the negative (but non-significant) correlation between weekly malaria cases and maximum temperature at shorter lags seen in hot districts may be due to its inhibitory and lethal effect on the survival of the parasites in the mosquitoes [ 30 ]. The survival rate of Anopheles gambiae is also reduced at higher temperatures. Nonetheless, the maximum temperature is not very extreme even in the relatively hot districts, thus the negative effect is not significant. As with all observational studies of malaria incidence and weather, a limitation of this study is the likely presence of some confounding factors that may have influenced the number of malaria cases and may have been associated with weather. Existing interventions such as insecticide residual spraying and other methods are routinely applied and were not included in this analysis. The results would have been biased by such confounding factors if interventions were undertaken on the basis of weather, or if they were undertaken on the basis of incidence and their effect was differential depending on the weather. Another minor problem is with the assumption of a finite length for the delayed effect of the different weather factors. However, the lag length was chosen based on the inter-relationship between weather, mosquitoes and parasites (Table 1 ) and it was assumed that this is biologically plausible. Weather factors alone explain seasonal cycles but were not accurate in explaining the magnitude of unusually bad years (Figure 3 ). This study was a scientific not a predictive exercise and suggests that no other factor is required for explaining seasonal cycles. A good early warning system has not been created, but some principles have been suggested for one. The first principle is that the lag length from time of rain to the expectation of malaria cases varies with climatic zone (with a saturating effect at higher rainfall levels), and rainfall may not be a key factor in urban malaria transmission. Secondly, minimum temperature is only important in the cold climatic zones, but not in the hot. Finally, maximum temperature makes little difference in either climatic zone. These key points need to be considered in the development of an early warning system for malaria. Such an early warning system would also include autoregressive terms or other terms that could improve prediction, but would have complicated the interpretation of coefficients in a model of the sort used, which was designed to detect the effects of weather factors on cases, rather than to predict case numbers. Such an early warning system is now being evaluated. Conclusions The findings are largely consistent with hypotheses, based on experimental data on mosquito and parasite development, about the interactions of climatic factors in determining the strength and lag structure of weather effects on falciparum malaria incidence. In the examined Ethiopian districts, weather-based predictors of malaria incidence are more useful in rural than in urban settings. These key points should be considered in the development of an early warning system for malaria. Authors' contributions HDT and ML conceived the study, undertook statistical analysis and drafted the manuscript. AT initiated the study and made data available in collaboration with WHO and Ministry of Health of Ethiopia. JS made major contributions to the study design and statistical analysis. All authors contributed to the writing of the manuscript and approved the submitted version of the manuscript. Supplementary Material Additional File 1 Three dimensional relationships between weather, lags and magnitude of effect at each lag Click here for file	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC535540.xml
535542	Saposin C promotes survival and prevents apoptosis via PI3K/Akt-dependent pathway in prostate cancer cells	Background In addition to androgens, growth factors are also implicated in the development and neoplastic growth of the prostate gland. Prosaposin is a potent neurotrophic molecule. Homozygous inactivation of prosaposin in mice has led to the development of a number of abnormalities in the male reproductive system, including atrophy of the prostate gland and inactivation of mitogen-activated protein kinase (MAPK) and Akt in prostate epithelial cells. We have recently reported that prosaposin is expressed at a higher level by androgen-independent (AI) prostate cancer cells as compared to androgen-sensitive prostate cancer cells or normal prostate epithelial and stromal cells. In addition, we have demonstrated that a synthetic peptide (prosaptide TX14A), derived from the trophic sequence of the saposin C domain of prosaposin, stimulated cell proliferation, migration and invasion and activated the MAPK signaling pathway in prostate cancer cells. The biological significances of saposin C and prosaposin in prostate cancer are not known. Results Here, we report that saposin C, in a cell type-specific and dose-dependent manner, acts as a survival factor, activates the Akt-signaling pathway, down-modulates caspase-3, -7, and -9 expression and/or activity, and decreases the cleaved nuclear substrate of caspase-3 in prostate cancer cells under serum-starvation stress. In addition, prosaptide TX14A, saposin C, or prosaposin decreased the growth-inhibitory effect, caspase-3/7 activity, and apoptotic cell death induced by etoposide. We also discovered that saposin C activates the p42/44 MAP kinase pathway in a pertussis toxin-sensitive and phosphatidylinositol 3-kinase (PI3K) /Akt-dependent manner in prostate cancer cells. Our data also show that the anti-apoptotic activity of saposin C is at least partially mediated via PI3K/Akt signaling pathway. Conclusion We postulate that as a mitogenic, survival, and anti-apoptotic factor for prostate cancer cells, saposin C or prosaposin may contribute to prostate carcinogenesis at its early androgen-dependent or metastatic AI state.	Background Androgens, growth factors, neuropeptides, and other trophic agents are involved in normal and neoplastic growth of the prostate. Prosaposin is the intracellular precursor of four lysosomal glycoproteins, saposins A-D, that are involved in lysosomal hydrolysis of sphingolipids. These saposins, through their interaction with glycosphingolipid hydrolases and their substrates, increase lysosomal hydrolytic activities. Saposins and prosaposin are expressed by various cell types and as a secretory protein in body fluids including blood, seminal plasma, seminiferous tubular fluid, and prostatic secretions [ 1 - 5 ]. Prosaposin and its active domain, saposin C, are known for their potent neurotrophic activities and are involved in neuro-embryological development [ 6 , 7 ]. The neurotrophic activity of prosaposin has been attributed to the NH 2 -terminal portion of the saposin C domain of the molecule which is the source for a number of biologically active synthetic peptides such as prosaptides TX14A [ 4 - 6 ]. Prosaptides (i.e., TX14A), saposin C, and prosaposin exert their biological effects by binding to a partially characterized single high-affinity G-protein coupled receptor (GPCR) [ 6 - 8 ]. It has been reported that mice with an inactivated prosaposin gene die at 35–40 days of age due to neurological disorders. These mice also develop several abnormalities in their reproductive organs, such as atrophy and involution of the prostate gland and inactivation of MAPK and Akt in the prostate epithelium [ 9 , 10 ]. The spectrum of biological activities of prosaposin or saposin C in cancer biology in general and in prostate cancer has not been specifically addressed. We have recently reported a higher expression of prosaposin in androgen-independent (AI) prostate cancer cells (PC-3 and DU-145) than in androgen-sensitive (AS) LNCaP or in normal prostate epithelial and stromal cells. In addition, we have found that prosaptide TX14A stimulates prostate cancer cell proliferation, migration, and invasion, activates the Raf-MEK-ERK-Elk-1 signaling cascade of the mitogen-activated protein kinase (MAPK) pathway, and inhibits the growth-inhibitory effects of sodium selenite administered at apoptogenic concentrations [ 11 ]. In the present study, we show for the first time that saposin C also functions as a survival factor, activates PI3K/Akt-signaling pathway, and in a cell type-specific manner, modulates the expression of procaspase- and caspase-3, -7, and -9 in prostate cancer cells under serum-starvation stress. We demonstrated that prosaptide TX14A, saposin C, or prosaposin decreased the growth-inhibitory effects, caspase-3/7 enzymatic activity, and apoptotic cell death induced by etoposide. In addition, our data show that saposin C activation of a p42/44 MAPK in prostate cancer cells is not only pertussis toxin-sensitive, but also PI3K/Akt-dependent. Moreover, the PI3K-inhibitor, LY294002, restores the apoptogenic effect of etoposide in prostate cancer cells studied. We propose that as a survival and anti-apoptotic factor, saposin C or prosaposin may contribute to prostate carcinogenesis or to the development of hormone-refractory prostate cancer. Results Saposin C acts as a survival factor for prostate cancer cells The effect of saposin C as a survival factor was assessed under serum-starvation stress. Androgen-sensitive (AS) LNCaP cells did not maintain their viability when cultured in serum-free, 0.25% FBS-RPMI, or 0.5% FBS-RPMI media, for more than 36 h. These cells started to detach from tissue culture plates and cell viability was decreased to less than 40% as determined by the trypan blue dye-exclusion method. However, in the presence of 1% FBS, cells remained attached to the tissue culture plate and their growth increased 31% at day 4 and 20% on day 6 as compared with the control values at day 2 (Fig. 1 ). Saposin C stimulated proliferation of these cells by 13% at day 2, 35% at day 4, and 33% at day 6 compared to the controls. PC-3 cells appeared to be more sensitive to serum deprivation and the number of live cells decreased 30% by day 4 and 60% by day 6 compared to the control values at day 2. However, saposin C (at 1.0 nM) increased cell proliferation by 9% at day 2, 19% at day 4, and 88% at day 6, compared to control plates at the same time period. The growth-response of DU-145 cells was different from PC-3 or LNCaP cells. In the absence of saposin C, the number of live cells increased 10% at day 4 and 29% at day 6 compared to day 2. These cells also demonstrated the highest proliferative response to saposin C at day 4 by 93% (Fig. 1 ). Taken together, these data indicate that saposin C in a dose-dependent and cell type-specific manner, promotes the survival of the serum-deprived prostate cancer cells. Figure 1 Saposin C acts as a survival factor for prostate cancer cells. Cells were cultured in their complete media for 3 days and shifted to their basal (serum-free) media or RPMI-1% FBS (only for LNCaP) in the presence or absence of the indicated concentrations of saposin C for 2, 4, or 6 days. Tissue culture media and saposin C were refreshed every 2 days. At the end of incubation periods, cells were trypsinized and cell number was determined using a hemocytometer and trypan blue exclusion method. PC-3 and DU-145 were used as androgen-independent and LNCaP cells were used as androgen-sensitive prostate cancer cell lines. Data represent the average of three independent experiments in triplicate samples; bars , ± SEM. * indicates P < 0.05, and ** indicates P < 0.01 compared to control. Statistical significance was determined by one-way ANOVA with Bonferroni's corrections. Saposin C activates the PI3K/Akt signaling pathway in prostate cancer cells Several studies have demonstrated that the serine/threonine kinase Akt is a pivotal survival effector for prostate cancer cells and protects them from apoptotic-cell death induction by various types of stresses. Hence, we next evaluated the effect of saposin C on the Akt signaling pathway in cells. Direct immunoblotting of serum-starved cells for 24 h showed that saposin C upregulates phosphorylative activity of Akt at serine 473 in androgen-independent (AI) PC-3 and DU-145 cells (Fig. 2A ). This response was biphasic. The response of LNCaP cells was distinct and started at 1.0 nM that subsequently returned to a basal level at higher treatment concentrations. Under our experimental conditions, we did not detect any changes in the phosphorylative activity of Akt at threonine 308. Since Akt is a downstream effector of PI3K, we tested the effect of the PI3K-specific inhibitor, LY294002. Pretreatment of cells with LY294002 (50 μM, 3 h) followed by saposin C treatment substantially reduced phosphorylation levels of both serine and threonine residues of Akt in AI- and AD-prostate cancer cells (Fig. 2A ). Figure 2 Saposin C activates Akt signaling pathway in prostate cancer cells. A , cells were cultured up to 70% confluency in their complete media, serum-deprived for 24 h, and treated with 10% FBS or saposin C at 0.1, 1, or 10 nM for 10 min. A representative culture plate was also treated with LY294002 (LY; 50 μM) before treating with saposin C (at 1 nM for LNCaP and 10 nM for PC-3 and DU-145). Fifteen μg protein per sample was subjected to SDS-PAGE under reducing conditions and immunoblotting was carried out using phospho-specific Akt antibodies against serine 473 or threonine 308. B , non-radioactive in vitro kinase assay was performed to determine the effect of saposin C on Akt kinase activity as described in details in Materials and Methods. Briefly, cells were grown as described above and Akt was selectively immunoprecipitated from 250 μg protein using 20 μl of immobilized Akt 1G1 monoclonal antibody. Immunocomplexes were pelletted and resuspended in kinase buffer in the presence of 200 μM ATP and 1 μg of Akt/PKB substrate-glycogen synthase kinase fusion protein (GSK-3α/β) and incubated for 30 min at 30°C, allowing immunoprecipitated Akt (if activated) to phosphorylate GSK-3. After terminating the kinase reaction, phosphorylated GSK-3 was detected by SDS-PAGE and immunoblotting using phospho-GSK-3α/β antibody. Control loading was evaluated with anti-Akt antibody to determine total Akt-level. Each experiment was performed in duplicate, and the assays were repeated three times. To determine whether the upregulation of Akt-phosphorylation by saposin C is associated with its kinase activity, in vitro kinase assays were performed. After 5 or 10 min exposure of cells to saposin C, activated-Akt induced phosphorylation of glycogen synthetase kinase-3 (GSK-3; a well-characterized Akt substrate) in both AS and AI prostate cancer cells (up to 3-fold compared to basal levels) at concentrations as low as 0.1 nM, was followed by a slight increase at higher concentrations (Fig. 2B ). Using purified human milk prosaposin, we observed similar responses (data not shown). The above results indicate that saposin C activates the Akt-signaling pathway in a PI3K-dependent manner in both AS and AI prostate cancer cells. Saposin C differentially modulates the expression or activity of caspases and PARP in prostate cancer cells under serum-starvation stress An essential component of the programmed death pathway in many cell types involves proteolytic cleavage of inactive caspases to catalytically active products. We investigated the expression of cleaved (active) and non-cleaved (inactive) forms of the initiator caspase-9, its active downstream effectors (caspases-3 and -7), and poly (ADP-ribose) polymerase (PARP, a nuclear substrate for caspase-3) in the cells after a 48 h serum deprivation period. Caspase-9 is closely coupled to proapoptotic signals and we found that the expression of procaspase-9 was not affected by saposin C; however, we were able to detect a reduction in its cleaved form at 10 nM saposin C in all cells investigated (Fig. 3 ). Figure 3 Effect of saposin C on expression/activity of caspases and PARP under serum-starvation stress. Cells were cultured routinely up to 60% confluency, washed with PBS, and incubated in their respective serum-free media supplemented with or without saposin C for 48 h. Cell lysates were prepared as described in Materials and Methods and 75 μg of clarified protein samples was subjected to SDS-PAGE under reducing conditions. Western analysis was carried out using monoclonal antibodies against non-cleaved and cleaved caspases-3, -7, and -9 and PARP. For control loading, membranes were probed or reprobed with anti-actin antibody. Each experiment was performed in duplicate, and the assays were repeated three times. With respect to the effector caspases, we noticed a dramatic dose-dependent increase in the expression of procaspase-3 in the AI PC-3 and DU-145 cell lines. However, we observed a reduction in expression of caspase-3 in both AS and AI cancer cells. Furthermore, our data showed that procaspase-7 expression in these cells was not affected by saposin C and under our experimental conditions we did not detect caspase-7 in AI prostate cancer cells. In LNCaP cells, we found a reduction in the level of caspase-7 at 1 or 10 nM of saposin C (Fig. 3 ). To further follow the mechanistic response of cells to saposin C in the death cascade, we examined the expression of one of the final death substrates, PARP, and its cleaved product. The intensity of PARP expression was considerably higher in PC-3 and DU-145 cells than in LNCaP cells. Saposin C, in a dose-dependent manner, increased PARP expression and this effect was associated with a parallel dose-dependent reduction of the cleaved (active) PARP levels (Fig. 3 ). Interestingly, the ratio of PARP: cleaved PARP expression in AI PC-3 and DU-145 cells, either at its basal level or after stimulation with saposin C, was higher than AS LNCaP cells. In general, saposin C induced a cell type-specific (AI versus AS) alteration in the expression level of initiator and effector caspases. This effect suggests a better survival and anti-apoptotic activity of saposin C in AI prostate cancer cells than in AS LNCaP cells. Saposin C protects prostate cancer cells from etoposide-induced apoptotic cell death Next, we decided to evaluate the effect of an apoptogenic agent, etoposide, on cell growth, apoptosis, and caspase activity in the presence or absence of various effectors. Cells were treated in complete culture media for three days, and then subjected to the MTS assay. Using these experimental conditions, we empirically determined the lowest concentration of etoposide that would lead to the highest growth inhibition. We found that the growth inhibitory effect of etoposide on prostate cancer cells is also cell type-specific. For example, a 20 μM etoposide concentration was sufficient to reduce the cell number to 53% in PC-3 and to 58% in LNCaP cells as compared to their control values. However, DU-145 cells were more sensitive and treating these cells with only 2 μM etoposide led to a 69% reduction in the cell number compared to control values (Fig. 4A ). Compared to etoposide-treated cells, saposin C increased cell growth by 13% in PC-3, 24% in DU-145, and 27% in LNCaP cells. Like saposin C, prosaposin reduced etoposide-induced growth inhibition to relatively the same degree. The highest increase in cell number was achieved with synthetic peptide TX14A treatment; however treatment of cells with the mutant 769M peptide showed only a negligible effect (2–5% increase the cell number) (Fig. 4A ). These results indicate that TX14A peptide, saposin C, or prosaposin can reduce etopside growth-inhibition on prostate cancer cells. Figure 4 Saposin C reduces growth inhibitory effect of etoposide and acts as an anti-apoptotic factor for prostate cancer cells. A , cells were seeded at 2000 per well in 96-well plates in their complete culture media for 3 (for PC-3 and DU-145) or 4 days (for LNCaP), treated with vehicle (DMSO), saposin C (0.1, 1, or 10 nM), prosaptide TX14A (10 nM), inactive mutant peptide 769M (10 nM), or prosaposin (1 ng/ml) at the indicated concentrations in the presence or absence of etoposide at the indicated concentrations for 3 days. After this period cell number was determined using MTS assay and cell type-specific OD/cell number calibration curve as described in Materials and Methods. B , apoptosis was determined by TUNEL assay. Cells were cultured in multiwell chamber slide up to 40% confluency in their complete culture media, and treated with etoposide in the presence or absence of saposin C (0.1, 1, or 10 nM) for 3 days. Percentage of apoptosis was determined by random selection of 10 microscopic field (at × 200 magnification) and cell count with a hemocytometer. Data expressed at the average of three independent experiments and twelve replicate samples; bars , ± SEM. * indicates P < 0.05, and ** indicates P < 0.01 compared to control (etoposide). Statistical significance of the effect of saposin C on cell growth and apoptosis was evaluated by one-way ANOVA with Bonferroni's corrections. Differences of vehicle (or etoposide)-only treated cells and any other single experimental group of interest (TX14A or prosaposin) was evaluated by Student's t -test and statistical significance was set at P < 0.05. Using the TUNEL assay and above experimental conditions, we next evaluated the effect of saposin C on the percentage of apoptotic cells after treating cells with etoposide for three days. Apoptotic cells were identified as dense, bright, and punctate, with brownish pigmentation of poly-fragmented nuclei. Among the three cell lines investigated, PC-3 proved to be the most resistant cell line to apoptosis induction by etoposide. Overall, there was a dose-dependent reduction (with a peak effect at 10 nM) of apoptotic cells in the three cell lines investigated. Saposin C decreased apoptotic cells by 47% in PC-3, 89% in DU-145, and 58% in LNCaP cells (Fig. 4B ). This result demonstrates the counteracting influence of saposin C and etoposide on apoptosis in prostate cancer cells. We also employed a sensitive fluorometric assay to measure caspase-3/7 (based on DEVDase) activity using the experimental conditions described above. Saposin C, at 1 nM concentration, demonstrated the highest reduction in caspase-3/7 activity in AS LNCaP (21%) and in AI PC-3 (35%) and DU-145 cells (30%) (Fig. 5A ). Prosaposin-treated cells also demonstrated a similar effect. TX14A peptide not only decreased the growth-inhibitory effect of etoposide (data not shown), but also proved to be a potent anti-apoptotic peptide, reducing caspase-3/7 activity by 43% in PC-3, 36% in DU-145, and 30% in LNCaP cells. However, the control (inactive mutant) peptide's (769M) effect was minimal (with a 2–5% reduction) (Fig. 5A ). These results clearly indicate that the anti-apoptotic activity of saposin C is at least partially associated with modulation of caspase-activity. Figure 5 Effects of saposin C on caspase-3/7 activity (A) and the influence of PI3-kinase inhibitor (B) in etoposide-treated cells. Cell culture, treatment period, growth and caspase activity was described in Figure 4. Caspase-3/7 activity was determined using the Apo-ONE Homoheneous Caspase-3/7 assay kit based on the cleavage of a profluorescent caspase-3/7 substrate (Z-DEVD-R110) and fluorimetric quantitation was performed at an excitation and emission wavelength of 485+20 and 535+25 nm, respectively. After correction of the fluorimetric reading with the blank (vehicle control), final fluorescent intensity was depicted as an arbitrary endpoint relative fluorescent unit, RFLU. PI3-kinase inhibitor (LY294002; LY) was used at final 1.5 μM concentration and saposin C was added at optimal 1 nM (for PC-3 and LNCaP) or 10 nM (for DU-145) concenration. Etoposide (Et) was added at optimal 20 μM (for PC-3 and LNCaP) or 2 μM (for DU-145). Data expressed are the average of three independent experiments and twelve replicate samples; bars , ± SEM. * indicates P < 0.05, and ** indicates P < 0.01. Statistical significance of the effect of saposin C on cell growth and apoptosis was evaluated by one-way ANOVA with Bonferroni's corrections. Differences of vehicle (or etoposide)-only-treated cells and any other single experimental group of interest (TX14A or prosaposin) was evaluated by Student's t -test and statistical significance was set at P < 0.05. The PI3K/Akt inhibitor restores apoptogenic activity of etoposide in saposin C treated cells To determine whether or not saposin C anti-etoposide apoptotic activity is PI3-kinase dependent, the effect of PI3K/Akt inhibitor (LY294002) on caspase-3/7 activity in the cells was examined in the presence or absence of saposin C ± etoposide. Through our initial studies, using trypan blue exclusion and MTS assays, we found 1.5 μM of LY294002 was a non-toxic and tolerable dosage for experimental period (3 days). Compared to DMSO-treated cells, LY294002 slightly increased (up to 10%) the caspase-3/7 enzymatic activity in PC-3 and DU-145 and showed almost no change in LNCaP (Fig. 5B ). As described above, saposin C significantly decreased induction of caspase-3/7 activity by etoposide. Saposin C also reduced the induction of casapases activity by LY294002 under our experimental conditions. Addition of LY294002 to the cells treated with saposin C and etoposide increased caspase-3/7 enzymatic activity, but to a level below than etoposide-only treated cells (Fig. 5B ). These results indicate that antiapoptotic activity of saposin C and its effect on caspase activity is at least partially mediated via the PI3K/Akt signaling pathway. Saposin C activation of MAPK is pertussis toxin-sensitive and PI3K/Akt-dependent In neuro-glial derived cells, neurotrophic activity, cell-death protection and the activation of MAPK by prosaptides (i.e., TX14A), saposin C, or prosaposin are mediated by their binding to a pertussis toxin-sensitive GPCR [ 6 , 7 , 12 , 13 ]. Our previous data demonstrated that prostate cancer cells were differentially responsive to the TX14A peptide in a number of biofunctional assays [ 11 ]. Our current results indicate the presence of a sensitive and /or responsive receptor-ligand interaction that could be accountable for the subsequent activation of downstream signaling effectors in MAPK- and Akt-signal transduction pathways. In addition, there is also emerging data indicating that signaling proteins such as PI3K and Akt can also activate MAPK pathways [ 14 - 16 ]. We have previously demonstrated MAPK-activation by the TX14A peptide derived from a trophic sequence of saposin C [ 11 ]. To examine the involvement of GPCR in saposin C-activation of the Akt-signaling pathway as well as the possibility of MAPK and Akt cross-signaling initiated by saposin C, we evaluated the effect of saposin C on p42/44 MAP kinase activation in prostate cancer cells in the presence or absence of various inhibitors. Treatment of cells with saposin C increased the phosphorylative activity of p42/44 MAPK, which was substantially inhibited by pretreating cells with the specific MEK inhibitor (U0126; Fig. 6 ). We used 10% FBS treatment as an external positive control for induction of p42/44 activity in the cells. Saposin C treatment of cells pretreated with PT showed a modest (in LNCaP cells) to strong reduction (in AI prostate cancer cells) in the level of phospho-p42/44 MAPK. This result clearly indicates a cell type-specific PT-sensitivity and/or the potential involvement of one or more G-proteins in saposin C-activation of MAPK pathway in the cells. Figure 6 Saposin C activation of MAPK pathway is pertussis toxin-sensitive and PI3K/Akt-dependent. Cells were cultured up to 60% confluency in their maintenance media, washed with PBS, serum-deprived for 20 h, and then fresh basal culture media was added for an additional 4 h in the presence or absence of LY294002 (50 μM, 3 h), Wortmannin (10 μM, 15 min), U0126 (10 μM, 1.5 h), or pertussis toxin (200 ng/ml, 4 h). After pretreatment, saposin C (0.1 nM) was added directly to the cells and incubated for 5 min at 37°C. Cell lysates were prepared and 10 μg protein per sample was subjected to SDS-PAGE and immunoblotting using phospho-specific p42/44 MAPK antibody. For control loading, membranes were also probed or reprobed with p42/44 antibody to detect total p42/44 MAPK. Parallel tissue culture plates, treated in the same manner, were also tested for cell viability by trypan blue dye-exclusion assay. FBS at 10% final concentration was used as a positive control. Each experiment was performed in duplicate, and the assays were repeated three times. The intensity of reduction of p42/44 activity and its cell type-specific pattern was very similar to PT plus saposin C- or PT-treated values. Interestingly, we observed a more profound reduction in p42/44 phosphorylative activity in cells pretreated with LY294002 and then treated with saposin C. To rule out the potential cytotoxic effect of the inhibitors, viability of cells was also determined by trypan blue dye-exclusion. We observed essentially similar results using the other structurally and mechanistically different PI3-kinase inhibitor (wortmannin). This experiment revealed that at the end of the pre-treatment incubation period, cell viability was equal to or more than 95%. These results indicate that MAPK activation by saposin C is at least partially mediated by saposin C-regulated PI3K/Akt pathways in prostate cancer cells. This result also provides additional proof for simultaneous activation of multiple (inter-related) signal transduction pathways by saposin C. Discussion Induction of apoptosis by androgen-ablation therapy significantly reduces androgen-dependent (AD) prostate cancer cells, but fails to cure the majority of patients due to the presence of apoptosis-resistant cancer cells that are androgen-independent [ 17 ]. It is likely that the development of these cells is an adaptive response to hormonal therapy rather the overgrowth of resistant cells. In-depth understanding of apoptotic phenomena, identification of its intracellular components, and characterization of its extracellular effectors as inducers or inhibitors may contribute to therapeutic approaches for prostate cancer. The prosaposin knock-out mouse model has revealed a number of interesting findings specifically in the male reproductive organs. Among these are atrophy of the prostate gland, epididymis, and seminal vesicles. Microscopic evaluation of affected tissues also shows undifferentiated phenotypes in prostate ventral and dorsal lobules and atrophy of the tubuloalveolar glands and their epithelial cell lining [ 9 ]. These data are suggestive of a primary role for prosaposin in development of the prostate gland. In a variety of neuro-glial derived cells, synthetic peptides encompassing a trophic sequence of saposin C and/or prosaposin have been found to induce growth, survival, and/or differentiation, or to prevent apoptotic cell-death in vitro and in vivo [ 11 , 18 - 20 ]. For example, prosaptide-TX14A and prosaposin, in a dose- and time-dependent manner reduced apoptotic cell-death induction of primary Schwann cells cultured in low serum concentrations, and PI3K inhibitors (wortmannin or LY294002) blocked their anti-apoptotic effects [ 21 ]. Here, we found that under prolonged serum starvation (2 to 6 days), although the responses of androgen-sensitive (AS) LNCaP and AI PC-3 and DU-145 prostate cancer cells were different from each other, saposin C in a dose- and time-dependent manner, proved to increase proliferation and survival of both cell types (Fig. 1 ). Normal prostatic epithelial cells neither tolerated the basal medium nor responded to saposin C (data not shown). Like many other cancer cells, withdrawal of mitogens, growth factors, and other trophic factors by serum-starvation, serves as a potent stimulus and driving force activating different survival mechanisms that eventually lead to apoptotic cell-death in AD- and Al-prostate cancer cells [ 22 , 23 ]. Since the PI3-Kinase/Akt signaling pathway is known as a central cell-survival mechanism and an important mediator of survival signals driven by growth or trophic factors, we were also interested in examining the level of PI3K/Akt activity during serum starvation. Interestingly, we found that saposin C upregulated Akt-p473 Ser phosphorylative activity in the prostate cancer cells under investigation. This effect was substantially inhibited by LY294002, an inhibitor of the upstream Akt effector, PI3K (Fig. 2A ). In addition, using an in vitro kinase assay, we proved that saposin C induction of PI3K/Akt activity in cells was associated with increased phosphorylation of GSK3α/β, as a downstream key target of the Akt kinase (Fig. 2B ). Unlike Akt-p473 Ser , our study showed a considerably higher level of constitutively activated Akt-p308 Thr in all cells and remained unaffected by saposin C. Other studies have also supported a central role for the PI3K/Akt pathway as a dominant growth factor-induced survival pathway for prostate cancer cells [ 14 , 16 ]. Constitutive activation of the PI3K/Akt survival pathway has been described as a mechanism that enables endocrine-related breast, lung, and prostate cancer to become refractory to cytotoxic therapy [ 24 - 27 ]. Interestingly, immunohistochemical analyses have demonstrated a direct correlation between Akt phosphorylation and the Gleason's score in prostate cancer [ 28 ]. Our results provide evidence that implicate the involvement of PI3K/Akt activity in saposin C (growth factor)-induced survival of prostate cancer cells. With respect to prosaposin, immunohistochemical staining of the involuted and atrophied prostate tissues from homozygous knock-out mice also showed inactivation of the MAPK and Akt signaling pathways [ 9 , 10 ]. Apoptotic-death signal transduction pathways are not limited to PI3K/Akt and may involve multiple redundant physiological pathways. Several studies have shown the involvement of caspases in the apoptosis of the prostate gland in normal development or malignant conditions [ 29 , 30 ]. For example, immunohistochemical evaluation in castrated mice and rats showed the presence of (activated) caspases in prostate and a correlation between caspase-3 expression and the Gleason's grade of tumors [ 29 ]. In vitro studies also revealed that caspase-inhibitory mechanisms might be involved in metastasis of prostate cancer cells [ 15 ]. We found that saposin C, in a dose-dependent manner, increased procaspase-3 and PARP levels and decreased the cleaved form of caspase-9 and -3 and PARP (a caspase-3 substrate) in both AS and AI prostate cancer cells. PARP cleavage has been recognized as a sensitive marker of caspase-mediated apoptosis and its cleavage paralyzes the enzyme's ability to repair DNA strand breaks. Therefore, reduction of the PARP cleavage is a strong indicator for anti-apoptotic activity of saposin C. Although procaspase-7 expression was not affected in any of the cells investigated, its active form was reduced only in LNCaP cells and was not detected in AI prostate cancer cells (Fig. 3 ). This special pattern for alteration in the level of procaspase-3, its cleaved form, and PARP was coincident with saposin C-induced cell survival under serum-deprivation culture condition (Fig. 1 ). Such divergent regulation of caspase-3 and PARP has rarely been reported in prostate cancer cells [ 31 ]. However, it has been demonstrated frequently in the nervous system and therefore might represent a unique characteristic of prosaposin or saposin C as a neurotrophic molecule [ 32 ]. Next, we exposed cells to a universal apoptogenic agent, etoposide, and found that prosaposin or its active derivatives (saposin C or TX14A peptide), were able to decrease the growth-inhibitory effect of etoposide-treated prostate cancer cells (Fig. 4 ). TUNEL assay, as a direct measure of apoptotic death showed a dose-dependent reduction in the percentage of apoptotic cells by saposin C (Fig. 4B ). Under similar experimental conditions, we also showed that saposin C, prosaptide TX14A, or prosaposin reduce caspase-3/7 activity in cells treated with etoposide. This effect could be counteracted by administration of a PI3-kinase inhibitor (LY294002) (Fig. 5A and 5B ). These data are a clear indication that saposin C-inhibition of the apoptogenic activity of etoposide is at least partially dependent on the upstream Akt effector, PI3K (Fig. 5B ). Together, the above findings suggest that the two closely inter-connected cell survival/apoptotic pathways (PI3K/Akt and caspases) activated by saposin C or prosaposin might potentially synergize and provide a growth and survival advantage to both AD- and AI-prostate cancer cells. Induction of mitogenic, survival, and anti-apoptotic signals in physiological and pathological conditions may begin from a wide array of extracellular stimuli and receptors, including receptor tyrosine kinases (RTKs) and G-protein coupled receptors (GPCRs). From a historical point of view, considerable attention has been given to the role of RTKs and their cognate polypeptide ligands in prostate cancer. However, accumulating evidence support the involvement of lysophosphatidic acids, neurotransmitters, and neuropeptides such as bombesin and neurotensin through GPCR signaling in the initiation or progression of prostate cancer [ 32 , 33 ]. GPCRs also use pathways that are very similar to those utilized by RTKs to activate survival and anti-apoptotic signaling pathways such as the prototypic Raf-MEK-MAPK and PI3K/Akt signal transduction pathways and caspase cascades [ 32 , 34 - 37 ]. Here, we showed that saposin C, in a pertussis toxin-sensitive manner, activated p42/44 MAPK (Fig. 6 ). Our study demonstrated the involvement of GPCR as a responsible receptor system interacting with saposin C and therefore activating the subsequent signaling pathways. Due to the considerable importance of activation of MAPK signal transduction activation by saposin C-GPCR and activation of the Akt signaling pathways, we tested whether inhibition of PI3K/Akt could affect saposin C-induced p42/44 MAPK activation. Pretreatment of cells with LY294002 inhibited saposin C activation of p42/44 MAPK (Fig. 6 ). These data not only indicate cross-communication between MAPK- and PI3K/Akt-signaling pathways, but also might suggest that simultaneous activation of the two important signal transduction pathways by saposin C provide a potent cell survival and apoptotic-death protection program for prostate cancer cells. Conclusion Our data for the first time show that by activation of multiple inter-related signaling pathways (PI3K/Akt and MAPK) and cell type-specific modulation of expression or activity of caspases, saposin C and/or its precursor (prosaposin) serve as a survival and anti-apoptotic factor for both AS- and AI-prostate cancer cells. Elucidation of such intricate mechanisms could potentially provide a therapeutic option that combines cytotoxic therapy and inhibition of survival/anti-apoptotic signals for AI metastatic prostate cancer. Finally, our observations provide novel insights into the diversity of biological activities of prosaposin in prostate cancer cells. Methods Cell lines Androgen-independent (PC-3, DU-145) and -sensitive (LNCaP) prostate cancer cell lines were obtained from the American Type Culture Collection (Manassas, VA) and grown in defined media (PC-3 and DU-145 in DMEM-10% FBS and LNCaP in RPMI-1640-10% FBS supplemented with 1 mM sodium pyruvate, 10 mM HEPES). Purified recombinant human saposin C and prurified human milk-prosaposin were characterized and provided by Dr. K. Sandhoff (University of Bonn, Germany) and Dr. M. Hiraiwa (University of California, San Diego), respectively. Cell survival assays Cells were initially grown in 100 mm plates in their respective culture media for 3 days, and after washing with PBS were incubated in serum-free DMEM (PC-3 and DU-145) or RPMI-1% FBS (LNCaP) in the presence or absence of saposin C (at 0.1, 1, or 10 nM) for 2, 4, or 6 days. Saposin C and culture media were replaced every 2 days. At the end of the incubation periods, cells were trypsinized and cell number was determined using a hemocytometer and the trypan blue exclusion method. Western analysis Protein expression analysis was performed according to standard procedures [ 38 ]. Briefly, the cell extract was prepared by washing cell monolayers with cold-PBS, lysing the cells on ice for 15 min with lysis buffer (20 mM PIPES [pH 7.4], 150 mM NaCl, 1 mM EGTA, 1% Triton X-100, 1.5 mM MgC1 2 ) supplemented with a protease inhibitor cocktail (Roche Diagnostic, Inc., Indianapolis, IN) and 1 mM sodium orthovandate, plus sodium dodecyl sulfate (SDS) at a final concentration of 0.1%. The lysates were then centrifuged (15 min, 4°C, 16,000 × g), aliquoted, and stored at -70°C until use. Protein concentration was determined by BCA assay (PIERCE, Rockford, IL). Each experiment was repeated at least two times. For western analysis, membranes were blocked with 5% BSA in the rinse buffer (150 mM NaCl, 20 mM Tris, 0.1% Tween 20) for 1 h, washed in rinse buffer for 10 min, and then incubated with the respective primary antibody at the indicated concentrations (see below). The membranes were then washed and incubated with the appropriate horseradish peroxidase-conjugated secondary antibody (1:1000 dilution; Santa Cruz Biotechnology, Santa Cruz, CA) for 1 h at room temperature, washed for 10 min and four more cycles of 5 min, and treated with an enhanced chemiluminescence (ECL) detection system (Amersham, Piscataway, NJ). In some cases, when the signal was very weak or undetectable, we used ECL-plus (Amersham). (i) Effect of PI3K/Akt and MEK-inhibitors, or Pertussis Toxin on Saposin C Activation of p42/44 MAPK Cells were grown in their respective complete culture media for 2–3 days (up to 60% confluency), washed with PBS, incubated in their serum-free (basal) media for 20 h, and then fresh basal media was added to all plates for an additional 4 h. Various inhibitors [LY294002 (50 μM, 3 h), Wortmannin (10 μM, 15 min), U0126 (10 μM, 1.5 h), and Pertussis toxin (200 ng/ml, 4 h)] were added to the culture medium, just before treating cells with saposin C (at 0.1 nM, 5 min). We used 10% FBS as a positive control. Cells were lysed and 10 μg of clarified protein samples was subjected to SDS-PAGE under reducing conditions. Phospho-specific p42/44 antibody (1:1000; Cell signaling Technologies, Bedford, MA) was used as the primary antibody and as a loading control. Filters were also probed or reprobed with anti-p42/44 antibody. Additional tissue culture plates that had been treated with or without inhibitors were also tested for cell viability by trypan blue dye-exclusion assay. (ii) Saposin C and PI3K/Akt signaling pathway Cells were cultured up to 70% confluency in their complete media and after washing with PBS, they were serum-starved for 24 h, and then treated with 10% FBS or saposin C at 0.1, 1 or 10 nM for 10 min. A representative tissue culture plate was also pretreated with the PI3K-inhibitor (LY294002, 50 μM for 3 h) before treating cells with saposin C (at 1 nM for LNCaP and at 10 nM for PC-3 and DU-145 cells). After preparation of the cell lysate, 15 μg of protein per sample was subjected to SDS-PAGE under reducing conditions. Immunoblotting was performed using phospho-specific Akt antibodies against serine 473 or threonine 308 (Cell Signaling Technology). A loading control was evaluated with anti-Akt antibody. Each experiment was performed in duplicate, and the assays were repeated three times. Immunoprecipitation and in vitro Akt kinase activity assay A non-radioactive Akt kinase assay kit (Cell Signaling Technologies) was used to determine whether saposin C treatment of cells under serum-starvation stress would lead to Akt-activation. For Akt-kinase assays, cells were grown up to 70% confluency in their maintenance media, serum-starved for 24 h, and then treated for 5 or 10 min in the presence or absence of saposin C at 0.1, 1, or 10 nM. Cells were washed once with ice-cold PBS and harvested under nondenaturing conditions using 1X ice-cold cell lysis buffer (from the Kit) supplemented with 1 mM phenylmethylsulfonyl fluoride (PMSF) on ice for 5 (or 10) minutes. Akt was selectively immunoprecipitated from 250 μg protein (whole cell lysates) using 20 μl of immobilized Akt 1G1 monoclonal antibody, and then incubated with gentle rotation for 4 h at 4°C. Samples were then centrifuged briefly (30 sec, 2000 × g) and pellets were washed twice with 1X lysis buffer and once with 1X kinase buffer. Immunocomplexes (pellets) were resuspended in 40 μl 1X kinase buffer [composed of 25 mM Tris (pH 7.5), 5 mM β-glycerolphosphate, 2 mM DTT, 0.1 mM Na 3 VO 4 , and 10 mM MgC1 2 supplemented with 200 μM ATP and 1 μg glycogen synthase kinase-3 [GSK-3; a well characterized Akt/PKB substrate] of fusion protein (GSK-3α/β) and incubated 30 minutes at 30°C, allowing immunoprecipitated Akt (if activated) to phosphorylate GSK-3. The kinase reaction was terminated by adding 20 μl of 3 × SDS sample buffer. Phosphorylated GSK-3 was then detected by western analysis using phospho-GSK-3α/β (for Ser 21 of GSK-3α and ser 9 of GSK-3β) antibody. The above in vitro kinase assay is based on the fact that phosphorylated-Akt (active) regulates GSK-3α/β kinase activity via phosphorylation at ser 21/9. For control loading, 10 μg protein per sample from the same whole cell lysates were subjected to western analysis using monoclonal anti-Akt antibody or actin. Each experiment was performed in duplicate, and the assays were repeated three times. Apoptosis assays (i) Effect of saposin C on expression of caspases by western analysis Cells were cultured up to 60% confluency in their complete culture media. After washing with PBS, they were incubated with their respective serum-free media in the presence or absence of saposin C at 0.1, 1, or 10 nM for 48 h; whole cell lysates were prepared as described above. Clarified protein samples (75 μg) were subjected to SDS-PAGE under reducing conditions. Western analyses were carried out using monoclonal antibodies against non-cleaved and cleaved caspases-3, -7, and -9 and poly (ADP-ribose) polymerase (PARP) provided in an Apoptosis Sampler Kit (Cell Signaling Technology). Each experiment was performed in duplicate, and the assays were repeated three times. (ii) Fluorometric measurement of caspase-3/7 activity in cells treated with etoposide We next examined the effect of an apoptogenic agent, etoposide, on cell growth and caspase activity in the presence or absence of saposin C, prosaposin, prosaptide TX14A, or an analogous inactive mutant peptide (769M; 4). Cells were seeded at 1,000 per well in 96-well plates in their complete culture medium for 3 to 4 days. After this period, cells were treated in complete culture media for 3 days with etoposide (2, 20, or 200 μM) to find the lowest concentration that led to the highest growth inhibition, as measured by MTS assay (described above). Using the optimal cell type-specific etoposide concentration (20 μM for PC-3 and LNCaP and 2 μM for DU-145), cells were treated with the vehicle (DMSO), saposin C (0.1, 1, 10 nM), TX14A (10 nM), mutant peptide (769M, 10 nM), or prosaposin (at 1 ng/ml). Using the cell type-specific OD/cell number calibration curve as obtained by MTS assay (Promega, WI), cell number per well was determined for the above treatment conditions. Parallel tissue culture plates were also used to determine caspase-3/7 activity using the Apo-ONE™ Homogeneous Caspase-3/7 assay (Promega, WI). This assay provides a homogeneous Caspase-3/7 reagent (Promega, Technical Bulletin-TB295) which performs a dual function, by rapidly and efficiently permeabilizing the cultured cells and at the same time exposing the intracellular space to the profluorescent caspase-3/7 substrate, rhodamine 110 (Z-DEVD-R110). After cleavage and removal of the DEVD peptides by caspase-3/7 activity, the fluorescence in each well was quantitated at an excitation wavelength of 485 + 20 nM and an emission wavelength of 535 + 25 nM and after correction based on blank control (DMSO-treated cells at a concentration equal to what used for dissolving etoposide) or the homogeneous caspase-3/7 reagent. Final fluorescent intensity was depicted as endpoint relative fluorescent unit, RFLU. Using similar experimental conditions but as an independent study, the effect of LY294002 (a PI3-kinase inhibitor) was also evaluated on growth and caspase-3/7 activity of cells treated with saposin C ± etoposide. After initial studies to find the optimal concentration, we used a non-toxic tolerable dosage of 1.5 μM for LY294002. In addition, we chose the most effective (optimal) concentration of etoposide (20 μM for PC-3 and LNCaP and 2 μM for DU-145) and saposin C (1.0 nM for PC-3 and LNCaP and 10 nM for DU-145) for this study. (iii) Terminal deoxynucleotide transferase-mediated nick end-labeling (TUNEL) Cells were cultured in multiwell chamber slides and treated with etoposide in the presence or absence of saposin C at 0.1, 1, or 10 nM as indicated above. In situ determination of apoptosis by Terminal dUTP nick-end labeling (TUNEL) was performed using an ApopTag Peroxidase In Situ kit as recommended by the manufacturer (Chemicon International, Temecula, CA). The ApopTag Kit detects single- and double-stranded DNA breaks associated with apoptosis. Drug-induced DNA damage is not identified by the TUNEL assay unless it is coupled with the apoptotic response. Briefly, at the end of the incubation period, cells were fixed in 1% paraformaldehyde in PBS, pH 7.4 for 10 min at room temperature, washed with PBS-twice, and permeabilized in pre-cooled ethanol: acetic acid (2:1) for 5 min at -20°C. After washing twice in PBS, 5 min each time, endogenous peroxidase activity in the cells was quenched in 3% H 2 O 2 in PBS for 5 min at room temperature, incubated with terminal deoxynucleotidyl transferase (TdT enzyme) and then with peroxidase-conjugated anti-digoxigenin antibody. Nuclear staining of the apoptotic cells was detected by 3',3'-diaminobenzidine tetrahydrochloride dihydrate substrate, as recommended by the manufacturer. Cells were then counterstained in 0.5% (w/v) methyl green and slides were mounted under a glass coverslip in permount mounting medium. For control staining, the enzyme incubation step was deleted. Microscopic examination of cells was carried out using a phase contrast microscope. Cells were counted by choosing ten random fields and the percentages of apoptotic cells were determined. Apoptosis was indicated by the presence of apoptotic bodies, exhibiting brightly labeled punctuated nuclei. Statistical analyses For cell survival and other quantitative data, a one-way analysis of variance (ANOVA) was employed to evaluate the influence of one variable on multiple independent groups. Bonferroni's corrections were also applied whenever a significant group effect was observed. To compare a control group with a single experimental group of interest, we used the Student's t-test. For cell survival studies, each treatment concentration was examined three times and in triplicate samples. The effect of saposin C or other effectors in the presence of etoposide on cell growth, apoptosis, or caspase-3/7 activity was studied in twelve replicates and repeated three times. Statistical significance was set at p < 0.05 or 0.01. Statistical analyses were performed using GraphPad Prism version 3.00 for Windows (GraphPad Software, San Diego, CA). Authors' contributions Author 1 (T-JL) carried out experiments described in figure 4 and 5B . Author 2 and 3 (OS and RL) reviewed the manuscript and provided valuable comments in different sections of the manuscript. Author 4 (SK) conceived the study, designed all the experiments, performed experiments described in figures 1 , 2 , 3 , 5A , and 6 , carried out the statistical analysis, and drafted the paper and wrote the final version of the manuscript. All authors read and approved the final manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC535542.xml
539350	The effect of cigarette price increase on the cigarette consumption in Taiwan: evidence from the National Health Interview Surveys on cigarette consumption	Background This study uses cigarette price elasticity to evaluate the effect of a new excise tax increase on cigarette consumption and to investigate responses from various types of smokers. Methods Our sample consisted of current smokers between 17 and 69 years old interviewed during an annual face-to-face survey conducted by Taiwan National Health Research Institutes between 2000 to 2003. We used Ordinary Least Squares (OLS) procedure to estimate double logarithmic function of cigarette demand and cigarette price elasticity. Results In 2002, after Taiwan had enacted the new tax scheme, cigarette price elasticity in Taiwan was found to be -0.5274. The new tax scheme brought about an average annual 13.27 packs/person (10.5%) reduction in cigarette consumption. Using the cigarette price elasticity estimate from -0.309 in 2003, we calculated that if the Health and Welfare Tax were increased by another NT$ 3 per pack and cigarette producers shifted this increase to the consumers, cigarette consumption would be reduced by 2.47 packs/person (2.2%). The value of the estimated cigarette price elasticity is smaller than one, meaning that the tax will not only reduce cigarette consumption but it will also generate additional tax revenues. Male smokers who had no income or who smoked light cigarettes were found to be more responsive to changes in cigarette price. Conclusions An additional tax added to the cost of cigarettes would bring about a reduction in cigarette consumption and increased tax revenues. It would also help reduce incidents smoking-related illnesses. The additional tax revenues generated by the tax increase could be used to offset the current financial deficiency of Taiwan's National Health Insurance program and provide better public services.	Background One of the problems in controlling tobacco in Taiwan is that cigarette prices are lower in Taiwan than in most countries [ 1 , 2 ]. In India, smokers have to work 77 minutes to afford a pack of cigarettes, in Indonesia 62 mins, in China 56 mins, but in Taiwan they need only work 7 to 10 mins to afford a pack [ 1 ]. As long as the domestic cigarette price remains rather low, we probably will not see much of a decrease in the number of smokers. The smoking population increased to 4.5 million persons (total population 22,520,776) in 2002 [ 3 ]. One out of three adults smoked. Currently, due to illnesses and death associated with cigarette smoking, smokers currently account for approximately 20 billion NT dollar of extra medical expense annually and account for a 160 billion NT dollar loss in GDP (Gross Domestic Product)[ 4 ]. This economic burden in putting pressure on the government to further increase the existing Health and Welfare Tax on tobacco. Because high excise taxes on cigarettes have been found able to reduce cigarette consumption [ 5 - 8 ], such measures are becoming one of the most important means of controlling tobacco [ 9 - 11 ]. The new tax scheme enacted on 1 January 2002 in Taiwan resulted in NT $16.8 tax excise. This tax included the existing taxes for NT $11.8 and a NT $5 Health & Welfare Tax for a 20-pack of cigarettes. The government also levies 5% sales taxes. Under that tax scheme, the cigarette tax revenues account for 40% of the retail price, which is about NT $42.2. While 40% sounds high, it is actually lower than the taxes imposed on cigarettes in developed countries that have seen some success at lowering cigarette consumption. The government should take elasticity of demand into consideration when deciding whether to increase or add an excise tax levy. If there is a price elasticity below 1, a tax increase brings about a decline in consumption and an increase in total tax revenues. Since Hsieh, Hu and Lin (1999) found that figure to be -0.6 in Taiwan, it can be reasonable to assume that a tax increase would be more likely to reduce cigarette consumption more significantly in Taiwan than in other countries with lower price elasticities at least in the short-term and medium term [ 12 ]. Higher taxes would also generate higher total tax revenues. Taiwan's Tobacco and Wine Tax Law is currently under review in the Legislative Yuan of Taiwan. Some legislators are seeking to increase the Tobacco Health and Welfare Tax by NT $3 per pack, raising it from NT $5 to NT $8 per pack. Just as they have done in the past, cigarette sellers will probably shift the tax increase to the consumers letting them be responsible for the increase. The effect of the price increase on demand depends on cigarette price elasticity – the larger the elasticity, the larger the reduction in consumption. Therefore, an estimation of price elasticity for domestic cigarettes could be a very important indicator of the possible effect a "Tobacco Health and Welfare Tax" would have on cigarette consumption and could be used to adjust Tobacco Health and Welfare Tax accordingly. Price elasticity of cigarette in Taiwan has been mostly estimated using time-series data [ 12 , 13 ], though this method might overlook the impact of smuggled cigarettes on the price elasticity and overestimate the price elasticity. It might be more appropriate and useful to use cross-section data from the Health Interview Survey to estimate the effect of cigarette tax on cigarette consumption and to compare differences in cigarette price elasticity with various smoker characteristics. Methods This study uses data on current smokers from 17 to 69 years of age during years 2000 to 2003. First, the demand function of the current smokers was established. Respondents who answered "everyday" or "some of the days" to the question how often you smoke were classified as "current smokers". Then, we used a random sampling of how they consumed and how much they paid for it between 2000 to 2003 to calculate cigarette price elasticity. We then analyzed differences in cigarette price elasticity in smokers categorized according to gender, age, education, income standard, and how much they smoked. Demand function Cigarette demand function was estimated by OLS and expressed by double logarithm function. The estimated demand function we used was: ln Q ig = α 0 + β 1 ln P ig + γ 2 ln I ig (1) where lnQ ig is i'th current smokers' logarithm representing monthly amount smokers consumed per person in group g. Current smokers were categorized according to age, gender, education, monthly income, and amount smoked. lnP ig is i'th current smokers' logarithm representing cigarette price per pack (NT$) for smoking characteristics group g. lnI ig is i'th current smokers' logarithm representing income per capita (NT$) for various categories of smokers in group g. The α 0 , β 1 , and γ 2 are parameters to be estimated. In order to measure how price change might affect cigarette consumption, the determination of price elasticity was particularly important. Price elasticity of demand for cigarettes is defined as the percent change in consumption resulting from a price increase. Cigarette price elasticity of demand β 1 and income elasticity of demand γ 2 can be derived from logarithmically differentiation (1) according to price and income. Data collection Using an annual face-to-face survey on cigarette consumption from 2000 to 2003 by Taiwan National Health Research Institutes, we collected data on how many packs current smokers consumed, how much they paid for a pack of cigarettes, how much they earned per month and how much they spent on cigarettes per month. Current smokers were categorized into gender, age, education, income, and amount smoked. Calculations of cigarette price were based on the average retail price of the top 3 most consumed cigarettes, calculations of number of packs smoked per month were done by dividing the monthly cigarette consumption by the average retail price. Calculations of income were based on personal monthly income. Certain background characteristics are listed in Table 1 . More than 90% of the smokers were men; less than 10% women. The numbers of young smokers were rising at the time of the study. Young people between the ages of 17 and 24 years old made up 5.4% of the sample in 2000, while they made up 12.8% in 2003, a 1.4 percent increase. The number of elderly smokers above the age of 55 gradually declined from 15.5% in 2000 to 10.8% in 2003. People with higher educational backgrounds tended to smoke more. More than 60% of the current smokers had senior or junior high school educations between 2000 and 2002. Second only to smokers with senior high school degrees, the percentage of the smokers with college degree increased by 27.2% in 2003, accounting for almost 40% of all the smokers. Thirty-five to forty percent earned between NT $20,000 to NT $30,000 per month. Those who smoked less than one pack were defined as light smokers; 1~2 packs (2 packs excluded), medium smokers; and 2 packs and above, heavy smokers. The proportion of light smokers had gradually increased from 50% in 2000 to 60% in 2003, and the proportion of heavy smokers gradually decreased from 11.8% in 2000 to 6.5% in 2003, showing an overall tendency toward reducing consumption. Table 1 Background characteristics of the current smokers in Taiwan, 2000–2003 2000 2001 2002 2003 Characteristics No. % No. % No. % No. % Total 856 632 521 493 Gender Male 789 92.2 599 94.8 496 95.0 460 93.3 Female 67 7.8 33 5.2 25 5.0 33 6.7 Age 17–24 46 5.4 35 5.5 41 7.9 63 12.8 25–34 179 20.9 133 21.0 94 18.0 101 20.5 35–44 299 34.9 222 35.1 190 36.5 177 35.9 45–54 199 23.2 149 23.6 122 23.4 99 20.1 55- 133 15.5 93 14.7 74 14.2 53 10.8 Education College and above 168 19.6 128 20.3 107 20.5 134 27.2 Senior high school 317 37.0 245 38.8 195 37.4 196 39.8 Junior high school 216 25.2 153 24.2 136 26.1 98 19.9 Primary school or lower 155 18.1 106 16.8 83 15.9 65 13.2 Month income No income 98 11.4 93 14.7 63 12.1 75 15.2 <NT $20,000 146 17.1 94 14.9 94 18.0 84 17.0 NT $ 20,000–39,999 335 39.1 237 37.5 192 36.9 174 35.3 NT $ 40,000–59,999 181 21.1 144 22.8 117 22.5 105 21.3 ≥ NT $ 60,000 96 11.2 64 10.1 55 10.6 55 11.2 Smoking degree Light smokers 445 50.0 319 50.5 281 53.9 296 60.0 Medium smokers 310 36.2 250 39.6 200 38.4 165 33.5 Heavy smokers 101 11.8 63 10.0 40 7.7 32 6.5 Results Cigarette consumption, retail price and personal monthly income were used in equation (1), the OLS method, to calculate cigarette price and income elasticities. The overall cigarette price elasticity was negative, less than one, indicating that cigarette consumption or demand in Taiwan was inelastic during the study period (Table 2 ). Taken into consideration that cigarette price elasticity is inelastic and that reduction of the cigarette consumption is done with a strategy of raising domestic cigarette price, we speculate that cigarette prices need to be even higher, to lower consumption enough to have a clear, strong impact in improved public health outcomes. While income elasticities were not statistically different from 2000 to 2002, they did reach a statistically significant level (5%) in 2003. At that time, estimated income elasticity was positive, indicating that cigarettes were normal goods. A positive value normally means that demand for normal goods would increase as incomes rise. But the value could vary greatly among normal goods. Interestingly, in our study, we found that as incomes were rose, income elasticity became low, indicating that income increases in 2003 only had a slight effect on cigarette consumption. Table 2 The estimated overall cigarette price and income elasticities of the current smokers, 2000–2003 a 2000 2001 2002 2003 Price elasticity -0.3134 (-2.894)* -0.3684 (-3.482)* -0.5274 (-3.143)* -0.3090 (-1.531) Income elasticity 0.0069 (0.708) 0.0042 (0.504) -0.0012 (-0.125) 0.0320 (2.916)* a. t ratios are shown in parentheses. * p < 0.05. After Taiwan's accession to WTO in 2002, Taiwan's first Tobacco Health and Welfare Tax added NT $5 the price of cigarettes, resulting in an increase in cigarette retail price from NT $35.2 in 2001 to NT $42.2 in 2002, about an NT $7 increase. In 2001, smokers 15 years old or older consumed an average 126.52 packs/person [ 14 ]. In 2002, the cigarette price elasticity became -0.5274, meaning that this price increase caused a reduction of cigarette consumption by 13.27 packs/person (10.5% per person) and a reduction of about 0.235 billion packs in the total consumption. Meanwhile, new cigarette consumption in 2002 was reduced to 2 billion packs for the population at and above the age of 15 in 2001. Provided there was a NT $16.8 tax on every pack of cigarettes, the tax revenue for the government would be 33.6 billion NT dollar. However, the tax has been in force for two years, and a significant reduction in consumption has been shown to date. Cigarette taxes accounted for 40% of the retail price in 2002. Although forty percent sounds high, this proportion is still rather low when, as mentioned earlier, comparing it with the 66% or more of the retail price going for cigarettes prices in high-income countries (with the notable exception of the United States)[ 9 ]. Consequently, cigarette prices in Taiwan are well below those of many high-income countries, who have seen significant reductions in cigarette consumption. Those successes certainly impose a pressure on the government to implement another tax increase in the existing Health and Welfare Tax. Inter-party negotiations at the Legislative Yuan resulted in changes to Tobacco and Liquor Tax Law which led to a NT $3 rise in the health tax levied on cigarettes, from NT$5 to NT$8 per pack. The cigarette industry is likely to pass the tax increase in the form of higher prices on to the consumer. A price increase of 3 NT per pack in cigarette would reflect a consumption reduction of 2.47 packs per capita, totaling 44.7 million packs, 2.2% per capita, in cigarette consumption. Cigarette consumption would be reduced to 2 billion packs (based on a population count of 2003), and the government tax revenue would be 33.46 billion, including the Health and Welfare Tax of 16 billion. We estimate that with a Health and Welfare Tax increase from 5 NT to 8 NT per pack, there would be a 6 billion NT increase in revenue from the Health and Welfare Tax. In 2000 and 2001, the cigarette price elasticities of the current smokers were -0.3134 and -0.3684, respectively. In 2002, it had reached its maximum of -0.5274, indicating that as cigarette prices increased, so did the price elasticity. Consumers responded to the higher prices by cutting consumption. Then in 2003, after Taiwan's accession to WTO, cigarette price elasticity then lowered to -0.309, indicating that the effect of a cigarette price hike can diminish over time. Two sets of previous data concerning the cigarette consumption, retail price and personal monthly income of the current smokers obtained during 2000–2001 and during 2002–2003, were made available to us. Using the OLS method and the rise of cigarette price in 2002 as the baseline, we computed the cigarette price and income elasticities for these two sets of data. Table 3 shows the differences in the overall cigarette price elasticity of the current smokers between before and after Taiwan's accession to WTO in 2002: before -0.4062 and after -0.3352. Our result clearly indicated that smokers were more responsive to the price after the cigarette price rose in 2002–2003. Yet, -0.4062 and -0.3352 also revealed that there was no big change in cigarette price elasticity before and after Taiwan's accession to WTO. The reason for this might be that the rise of average cigarette price was only NT $7 per pack. We speculate that a possible continuous and substantial rise in cigarette prices in the future might increase the overall price elasticity, which in turn could allow for a more effective use of the tax increases or price increases to control tobacco. Table 3 The estimated cigarette price and income elasticities of different types of current smokers during 2000–2001 and during 2002–2003 a 2000–2001 2002–2003 Characteristics Price elasticity Income elasticity Price elasticity Income elasticity Overall -0.3352 (-4.355)* 0.0055 (0.849) -0.4062 (-3.102)* 0.0174 (2.331)* Gender Male -0.3107 (-4.036)* 0.0020 (0.291) -0.3930 (-2.935)* 0.0194 (2.450)* Female -0.1223 (-0.312) -0.0206 (-0.820) -0.1406 (-0.251) -0.0476 (-1.903) Age 17–24 -0.4022 (-0.645) 0.0368 (1.151) -0.1057 (-0.144) 0.0420 (1.532) 25–34 -0.0528 (-0.274) -0.0038 (-0.227) 0.2300 (0.734) 0.0315 (1.822) 35–44 -0.1835 (-1.425) -0.0028 (-0.206) -0.2154 (-1.023) -0.0060 (-0.422) 45–54 -0.1540 (-1.087) 0.0048 (0.388) -0.4100 (-1.624) -0.0107 (-0.724) 55- -0.2453 (-1.073) -0.0080 (-0.608) -0.0475 (-0.118) 0.0008 (0.044) Education College and above -0.1070 (-0.514) -0.0133 (-0.584) -0.7007 (-2.047)* 0.0919 (4.148)* Senior high school -0.2772 (-2.273)* 0.0211 (1.825) -0.5369 (-2.703)* 0.0233 (2.069)* Junior high school -0.5766 (-4.093)* 0.0275 (2.552)* 0.1789 (0.833) 0.0279 (2.058)* Preliminary or lower 0.0560 (0.302) -0.0056 (-0.440) -0.0392 (-0.118) -0.0191 (-1.244) Month income No income -0.5103 (-2.193)* -0.8363 (-2.014)* <NT $20,000 -0.4570 (-2.373)* -0.7478 (-2.316)* NT $ 20,000–39,999 -0.4805 (-3.953)* -0.2861 (-1.345) NT $ 40,000–59,999 -0.1562 (-0.951) -0.2624 (-1.022) ≥ NT $ 60,000 0.2341 (1.050) -0.1152 (-0.301) Smoking degree Light smokers -0.1984 (-2.046)* 0.0087 (1.120) -0.5320 (-3.293)* 0.0172 (1.956) Medium smokers -0.0228 (-0.914) -0.0005 (-0.198) -0.2600 (-5.068)* -0.0046 (-1.514) Heavy smokers 0.2573 (1.973)* -0.0083 (-0.820) -0.0006 (-0.005) 0.0010 (0.137) a. t ratios are shown in parentheses. * p < 0.05. Cigarette price elasticity for the male smokers reached statistically significance (5%), which was higher than the price elasticity of the female smokers, indicating that men were more responsive to price elasticity than the women. The cigarette price elasticity of the male smokers in 2002–2003 was -0.393, which was higher than -0.3107 male smokers in 2000–2001. The income elasticity of male smokers in 2002–2003 was 0.0194, reaching statistical significance (5%). There was no statistical difference found in estimated cigarette price and income elasticity of smokers among the different age sub-groups, though, according to some reports in foreign countries, teenagers are more sensitive changes to cigarette prices than adults [ 15 - 17 ]. Education level seemed to make a difference. The price elasticity smokers with junior high school educations was -0.5766 in 2000–2001, which was higher than that of senior high school level smokers, -0.2772. In 2002-2003, the price elasticity of college level smokers was -0.7007, which was higher than that of the smokers of senior high school level, -0.5369. Therefore, between 2000 and 2001, the more educated group had a larger price elasticity, though the less educated group had the higher coefficient in 2002–2003. Our findings cover two time spans and are different from those reported by foreign researchers who have found that consumers at lower education levels respond stronger to the change of cigarette price [ 18 ]. In our study, those with no income had the greatest cigarette price elasticity, -0.5103 in 2000–2001 and -0.8363 in 2002–2003. Those with no income were more sensitive to the price rise than those with income. Light smokers had the highest cigarette price elasticity, -0.1984 in 2000–2001 and -0.532 in 2002–2003. Heavy smokers had a price elasticity of 0.2573 in 2000–2001, indicating that with the rise of cigarette price, these smokers increased their consumption of tobacco. Conclusions In this study, we evaluate the effect of a new tax on the consumption of tobacco by calculating the cigarette price elasticities of various kinds of smokers and comparing those values with their reactions to an increase cigarette, and then using knowledge gained from that study, we assess the possible effect of another increase on consumption. We found that the new tax scheme implemented after Taiwan joined the WTO reduced cigarette consumption by 13.27 packs/person (10.5%). We estimated that an additional NT $3 increase in Taiwan's Health and Welfare Tax would reduce cigarette consumption here by 2.47 packs/person (2.2%). In this study, cigarette price elasticity was less than one, meaning that in addition to reducing cigarette consumption, an additional tax would also generate additional tax revenues. Continuing price increases should reduce cigarette consumption significantly. Provided that the tax increases are proportionately larger than the resulting reduction in cigarette consumption, cigarette tax revenues will increase and can be used to reduce current National Health Insurance deficits and possibly reduce the damage and death caused by smoking related diseases. Based on estimated cigarette price elasticities for various kinds of current smokers, we found that male smokers without income and light smokers were more sensitive to changes in cigarette prices. Teenagers (17 – 24 years old), however, were not found to be significantly influenced by the change in cigarette price, which means it will take more than just tax increases to decrease consumption among our youth. Schools will need to commit to preventive education by inculcating the students with the knowledge of tobacco hazards. Only through early preventive education starting from their childhood can we expect to see significant reduction in cigarette consumption. Finally, the R-squared statistics of the each empirical estimation was below 0.1 for probably sake of to calculate elasticity and exclude variables like advertisement. Future research should also attempt to include these factors with cigarette demand function. Competing interests The author(s) declare that they have no competing interests. Authors' contributions JML performed physical measurements, collected data, and drafted the manuscript. TCH reviewed the manuscript. CYY and SHC carried out the statistical analysis and participated in data collection. All authors read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC539350.xml
535556	The architecture of chicken chromosome territories changes during differentiation	Background Between cell divisions the chromatin fiber of each chromosome is restricted to a subvolume of the interphase cell nucleus called chromosome territory. The internal organization of these chromosome territories is still largely unknown. Results We compared the large-scale chromatin structure of chromosome territories between several hematopoietic chicken cell types at various differentiation stages. Chromosome territories were labeled by fluorescence in situ hybridization in structurally preserved nuclei, recorded by confocal microscopy and evaluated visually and by quantitative image analysis. Chromosome territories in multipotent myeloid precursor cells appeared homogeneously stained and compact. The inactive lysozyme gene as well as the centromere of the lysozyme gene harboring chromosome located to the interior of the chromosome territory. In further differentiated cell types such as myeloblasts, macrophages and erythroblasts chromosome territories appeared increasingly diffuse, disaggregating to separable substructures. The lysozyme gene, which is gradually activated during the differentiation to activated macrophages, as well as the centromere were relocated increasingly to more external positions. Conclusions Our results reveal a cell type specific constitution of chromosome territories. The data suggest that a repositioning of chromosomal loci during differentiation may be a consequence of general changes in chromosome territory morphology, not necessarily related to transcriptional changes.	Background It is a longstanding observation that chromatin distribution in the interphase cell nucleus varies with the cell type. Flemming described differences in nuclear appearance in 1882 [[ 1 ], p.100]. Since then methodological advancements have made it possible to study nuclear chromatin architecture in much more detail. The spatial restriction of each chromosome to a limited area of the interphase nucleus, the chromosome territory, has been unequivocally demonstrated by fluorescence in situ hybridization (FISH) [ 2 , 3 ]. However, although progress has been made over the last decade [for reviews see [ 4 - 7 ]], the internal organization of chromosome territories is still largely unknown. Here we asked whether chromosome territories display differences between cell types in their internal chromatin organization. We amended our experimental approach with the determination of the position of a gene locus relative to its chromosome territory. In several previous studies it was observed that a number of active genes located preferentially at the surface of their chromosome territories or even outside [ 8 - 12 ], while others noted that active genes could also be positioned in the chromosome territory interior [ 13 ]. A general labeling of transcription sites resulted in signals throughout chromosome territories [ 14 , 15 ] demonstrating that the periphery of chromosome territories is not the only region where transcription occurs. Fluorescence in situ hybridization (FISH) studies investigating the major histocompatibility complex MHC [ 10 ] or the epidermal differentiation complex EDC [ 11 ] showed looping beyond the surface of the chromosome territory upon activation in up to 25% of the cases. Both loci are in the megabase size range with multiple co-regulated genes. Even higher frequencies of a location outside the territory were described for a gene rich human region without coordinate gene expression on chromosome 11p15.5 [ 16 ] and for genes of the Hoxb cluster in mouse embryonic stem cells entering differentiation [ 12 ]. It has thus been suggested that strongly expressed genes may be on chromatin loops that loop to the periphery of the territory, while genes expressed at low levels may occupy either a more interior or a random position [ 13 ]. Difficult to interpret were data concerning looping out of the β-globin gene locus from its chromosome territory in mouse erythroleukemia cells. In unstimulated cells where the locus shows DNase-hypersensitive sites but is not yet expressed, nearly half of the loci looped out. In stimulated cells where expression occurs, however, this was found only in about a third of the cases [ 17 ]. Consistent with the looping out of endogenous loci found in FISH studies, an opening of GFP-labeled artificial chromosomal regions was observed upon transcriptional activation or binding of transcription factors [ 18 - 24 ]. So far, a correlation of gene activation with increasing looping-out from the chromosome territory has only been shown for gene clusters but not for single genes. In the present study we have chosen the chicken lysozyme gene ( cLys ), which is highly active in macrophages, as a model system to explore the possibility of positional changes during activation of a single gene. cLys does not have co-regulated neighbors. Recently the gene cGas41 was found only 200 bp downstream of the polyA-site of cLys . cGas41 is expressed on a low level in all chicken tissues and cell lines tested, including all cell lines used here [ 25 ]. Macrophage differentiation is an interesting model system for studies of cell fate decisions. As all blood cells, macrophages originate from pluripotent hematopoietic stem cells and develop via defined multipotent and then progressively restricted precursor types. The developmental regulation of lysozyme in this differentiation system is well characterized. Expression is not detectable in multipotent myeloid precursors, which are able to differentiate to either the erythroid, granulocytic or the macrophage lineage (Figure 1 ). The gene is also not expressed in the erythroid lineage. Expression is first detected at a low level in granulocyte-macrophage precursors (myeloblasts) and is further upregulated in macrophages. By the addition of bacterial lipopolysaccharide (LPS) to macrophages, another tenfold increase in lysozyme expression is caused [ 26 ]. Studies of cLys regulation were greatly facilitated by cell lines representing these differentiation stages [ 27 , 28 ]. Figure 1 Cell lines used in this study. Pluses and minuses indicate the expression state of the lysozyme gene. Colors are the same as used in Figures 4, 5 and 7. The chicken karyotype consists of several pairs of so called macrochromosomes with sizes comparable to that of mammalian chromosomes and many much smaller microchromosomes [for review see [ 29 , 30 ]]. cLys is located on the short arm of chromosome 1 which is with about 190 Mbp comparable in length to human chromosome 4 [ 31 ]. In the present study, we investigated the large-scale chromatin organization of chromosome territories in well characterized cell lines representing the five chicken cell types described above (Figure 1 ). Multi-color 3D FISH was applied to cells with structurally preserved nuclei, followed by confocal microscopy and three-dimensional image analysis. We assessed the morphology of chromosome territories 1 and 8 by visual inspection and measured the dispersion of the painted territories in each cell type. We demonstrate that chromosome territory dispersal increases in the more differentiated cell types. Further, we determined the 3D positions of the chicken lysozyme gene domain (including cGas41 ) and the chromosome 1 centromere relative to their chromosome 1 territory. We found that not only the lysozyme gene domains but also the centromeres were mostly in the chromosome territory interior in multipotent myeloid precursor cells and relocated to the territory periphery in further differentiated cell types. In addition, we determined the radial positioning of the chromosome territories 1 and 8 within the nuclei of each cell type and measured nuclear volumes. Results The morphology of chromosome territories changes during differentiation 3D FISH was performed on formaldehyde fixed, structurally preserved nuclei. Visual examination of painted chromosome 1 and chromosome 8 territories revealed differences between the cell types (Figure 2 , Figure 3 ). Territories in multipotent myeloid precursor cells were relatively compact and homogeneously stained (Figure 2b , Figure 3a ). In proerythroblasts, territories were more diffuse and borders became less definable (Figure 2f,2g , Figure 3d ). These changes cannot be explained with an increase in nuclear volume since nuclei of proerythroblasts were smaller than those of myeloblasts (see below and Figure 4 ). In a minority of proerythroblasts, in addition to the labeled territories our paint probe labeled DNA-clusters in the center of the nucleus. These clusters had low DNA counterstain but were often associated with strongly counterstained regions (see arrow in Figure 2g ). The clusters may play a role in forming heterochromatin as observed during differentiation of human erythroid cells [ 4 ]. Chromosome territories in myeloblasts (Figure 2c , Figure 3b ) appeared less compact than in precursors but more compact than in proerythroblasts. Territories in unstimulated macrophages (Figure 2d ) had even more diffuse borders. In their interior we observed agglomerations of labeled DNA and lacunas in some cases. A maximum of dispersal was noted in stimulated macrophages (Figure 2e , Figure 3d ). Here, territories had grooved, fuzzy surfaces and a heterogeneous label throughout. Lacunas were frequent in the larger chromosome 1 territories. Figure 2 FISH with a chromosome 1 paint probe (red) and the lysozyme gene domain probe (green/yellow). (a) FISH on metaphase chromosomes. The chicken lysozyme gene domain is located on the short arm of chromosome 1. Note that the library probe mix used gives particularly strong signals at the centromeres (arrows). (b-g) 3D-FISH on structurally preserved nuclei. For each cell type, single confocal sections of one nucleus are shown. In b-f, nuclear outlines were drawn after the DNA counterstain which was omitted from the figure to avoid obstruction of the territory signals. In addition to the cLys domain signals, centromeres are in focus in some of the sections (arrows). (b) Multipotent myeloid precursor cell. (c) Myeloblast. (d) Macrophage without LPS-activation. (e) Macrophage with LPS-activation. On the right hand side, a threshold of 80 was applied to the territory signal of the central image to visualize disaggregation into several objects. While usually only few objects are present in any given focal plane, in this particular example the breakup is well recognizable. The algorithm applied in the calculations works on 3D-stacks, however. The macrophage cell line is aneuploid (see Methods), the cells shown in d and e have three territories with chromosome 1 material, each containing a cLys signal. (f) Erythroblast. (g) An additional section of the erythroblast shown in f visualizes a cluster of chromosome 1 material (red in left image) in a central nuclear area (arrow) with low DNA-counterstain but associated with a brightly stained region (see main text). Such clusters were less pronounced when only paint probes from early DOP-PCR-amplification rounds were used (see Methods for details). DNA counterstain is blue in left, gray in right image. Scalebar 5 μm for b-g. Whereas in precursor cells the lysozyme gene domain signal was found nearly always inside the territory, in differentiated cells more external positions were frequent. Note that the multipotent myeloid precursor cell has a relatively small nucleus and nuclear volume is increased in the further differentiated cell types. Figure 3 3D-FISH on structurally preserved cell nuclei with paints for chromosome 1 (red) and chromosome 8 (green). For each cell type, two confocal sections of one nucleus are shown. Nuclear outlines were drawn after the DNA counterstain which was omitted to avoid obstruction of the territory signals. (a) Multipotent myeloid precursor. (b) Myeloblast. (c) Macrophage activated with LPS. (d) Proerythroblast. Scalebar 5 μm. Figure 4 Nuclear volumes. Each dot indicates the volume of one nucleus. Nuclei from experiments with hybridization of chromosome 1 and cLys probes (left) and those with chromosome 1 and 8 probes (right) were sorted by size to allow an easier comparison by eye. See main text for mean values. A comparison of chromosome territory surface and chromatin texture by visual inspection is bound to subjective influences. To allow an unbiased, quantitative evaluation of chromosome territory morphology, we counted the number of objects to which chromosome territories disaggregate at increasing threshold levels. With a computer program newly developed for this purpose (for details see Methods and Figure 2e ) we could confirm the visual impression of relatively compact chromosome 1 territories in multipotent precursors by showing that at higher thresholds they disaggregate to smaller numbers of objects than the fuzzier territories of more differentiated proerythroblasts (Figure 5a ). Statistical inferences about the means of the maximal number of objects in these three cell types were highly significant as determined by Analysis of Variance (one-way ANOVA; F(2,119) = 58.5, p < 0.001; see Methods for details). Post-hoc Sidak tests revealed a significant difference between precursor cells and myeloblasts (p = 0.013) and highly significant differences between proerythroblasts and the two other cell types (p < 0.001). Chromosome 1 territories in macrophages also yielded high numbers of objects. However, macrophages of the utilized cell line contain an additional fragment of the short arm of chromosome 1 and sometimes complete additional chromosomes 1. Due to this aneuploidy object numbers are biased towards higher numbers. Results are thus not directly comparable with other cell lines. Notably, in stimulated macrophages territories disaggregate into more objects than in unstimulated ones, suggesting that stimulation triggered a change to a more dispersed chromatin texture (p < 0.001). The stronger dispersion of chromosome 1 territories in more differentiated cells was confirmed in a second experimental series with painted chromosome 1 and chromosome 8 territories (Figure 3 ) in multipotent precursors, proerythroblasts, myeloblasts and activated macrophages (data not shown). The DNA content of chicken chromosome 8 is about 30 Mbp [ 31 ]. This is roughly one sixth of the DNA content of chicken chromosome 1 and about two thirds of the smallest human chromosome, 21. Chromosome 8 was diploid in all utilized cell lines. As expected due to its smaller size, it disaggregated in all cell types to a much smaller number of objects (Figure 5b ). ANOVA analysis of all groups showed a highly significant aberration from the assumption of similar distributions in all cell types (F(3,116) = 38.2, p < 0.001). A difference between multipotent precursors, proerythroblasts or myeloblasts was not detectable (p > 0.9 in post-hoc Sidak tests) but in activated macrophages chromosome 8 territories did break up to a larger number of objects than in other cell types over a wide threshold range (p < 0.001 with all other cell lines). Figure 5 Disaggregation of chromosome territories in objects. (a,b) mean number of objects at increasing thresholds for chicken chromosome 1 (a) and chromosome 8 (b) territories. When the starting threshold of 20 is gradually increased, the nuclear background produces at first few and then many objects (around threshold 40). Suppression of nuclear background occurs at thresholds between 60–70, leaving chromosome territories only. The range above these thresholds is thus the most interesting since it is here where the territories start to break up in several objects (compare Fig. 2). These objects are gradually lost at further increasing thresholds. Values for macrophages are not directly comparable to other cell types since they contain additional chromosomes (see Methods). (c, d) Chromatin content per surface. Signal intensity of objects was divided by object surface area and averaged (see Methods for details). Since additional chromosomal parts add intensity as well as surface, this parameter is unsusceptible to aneuploidy. To allow a comparison of chromosome 1 territories in the aneuploid macrophages with those of other cell types, we analyzed the intensity of objects, i.e. their chromatin content, per surface area (Figure 5c,5d , see Figure legend and Methods for details). The chromatin content per object surface area was measured in multipotent precursor cells for both, chromosomes 1 and 8, again confirming their more compact structure in this cell type as compared to more differentiated cells. As a third parameter, we measured the average amount of labeled chromatin (signal intensity) per voxel (volume pixel) of the segmented objects. This parameter did not show differences between the cell types. Thus in all cell types a given amount of chromatin within the segmented objects was distributed over a similar volume over a wide threshold range (data not shown). The positions of the lysozyme gene domain and of the chromosome 1 centromere change during myeloid differentiation To determine the positioning of the lysozyme gene domain relative to the chromosome 1 territory, we performed dual color FISH with a chromosome 1 paint probe and a 20 kb plasmid probe for the lysozyme gene domain (Figure 2 ). By using a particular probe mix (see Methods) we were able to obtain an especially bright signal at the centromere in the same color channel as the paint probe. The positions of both, the lysozyme gene domain and the centromere, differed largely between the cell lines representing the various differentiation stages. To classify the positions of the signals, we used the scheme shown in Figure 6a . In multipotent precursor cells (Figure 2b ) we found the cLys gene domain inside the harboring chromosome territory, away from the territory border (categories A and B) in 48% of the cases (Figure 6b ). In additional 46% the gene signal was inside the territory touching the border (cat. C). It was previously shown that these cells do not express lysozyme but do show low level expression of the neighboring cGas41 [ 25 ]. We conclude that a location inside the territory is compatible with low-level expression. In further differentiated cells, the lysozyme gene locus was found more often in the periphery of chromosome 1 territories (Figure 2 , Figure 6b ). This is true for myeloblast/macrophage lineage cells with lysozyme expression as well as for proerythroblasts in which the expression of cLys and cGas41 does not differ from the precursor cells. The most peripheral localization, sometimes outside of the painted territory, was found in activated macrophages (Figure 2e ) where the cLys expression level is highest. Here 82% of the gene signals were on the surface or further outside (cat. D-F). The difference in distribution between precursor cells and all other cell types was highly significant (p < 0.001) as was the difference between activated macrophages and all other cell types (p < 0.001). The difference between unstimulated macrophages and proerythroblasts (p = 0.003) or myeloblasts (p = 0.024) was also significant whereas the difference between myeloblasts and proerythropblasts was not (p = 0.5). To test the robustness of our results, we repeated statistical analysis after reducing the number of applied categories of localization to only three: internal (A+B), peripheral (C-E) and external (F+G). We confirmed highly significant differences when precursor cells or activated macrophages were compared to any other cell type (p = 0.003 or smaller). In summary, for the lysozyme gene we found a change in position from interior when not expressed in myeloid precursor cells to peripheral when strongly expressed in activated macrophages. This would fit the hypothesis that highly expressed genes are preferentially located in the territory periphery, as it was found previously for large gene clusters [ 10 , 11 ]. However, this hypothesis does not explain the difference in positioning between the precursors and the proerythroblasts. Figure 6 Classification of cLys gene domain and centromere signals. (a) Scheme used to classify the localization of gene and centromere signals relative to their chromosome 1 territory [adopted from 11]. The red ellipsoid represents the territory, the yellow dots the signals of genes or centromeres. Categories are: A, inside the territory delineated by the paint probe, away from the surface. B, inside, closer to the territory surface but not touching it. C, inside and touching the surface. D, on the surface. E, outside and touching the surface. F, without contact to the territory but in immediate neighborhood. G, away from the territory. (b, c) Distribution of the lysozyme gene domain (b) and the centromere (c) relative to the chromosome 1 territory in 5 different cell types. Between 79 and 95 cLys gene domain and centromere signals were evaluated for each cell line and assigned to the categories A-G. Surprisingly, the centromeres of chromosome 1 showed a change in distribution very similar to the cLys gene domain (Figure 6c ). In no cell type we found a significant difference between the two (p = 0.255 or larger). For example, in multipotent precursors all detected centromeres were inside the territory, either without (cat. A, B) or with contact to the surface (cat. C). In contrast, in activated macrophages 93% of the centromeres were on the surface (cat. D) or outside with contact to the surface. Again, the difference in distribution between precursor cells and all other cell types was highly significant (p < 0.001) as was the difference between activated macrophages and all other cell types (p < 0.001). Myeloblasts and unstimulated macrophages showed a moderately significant difference (p = 0.044) whereas the differences between proerythroblasts and myeloblasts (p = 0.654) or unstimulated macrophages (p = 0.084) were not significant. When applying only three categories of localization as described above, differences between precursor cells or activated macrophages and any other cell type again where highly significant (p = 0.001 or smaller) with the exception of precursor cells compared to myeloblasts showing a moderate significant difference (p = 0.035). Silent lysozyme genes do not colocalize with centromeric heterochromatin Brown et al. [ 32 , 33 ] showed examples of genes in hematopoietic cell types, which were tethered to centromeric heterochromatin when silent, but located remote from heterochromatin when active [ 4 , 32 , 34 ]. We asked whether the same nuclear location could be found for silent and active lysozyme genes. A probe that would label all centromeres of chicken chromosomes is not available. We reasoned that if the inactive lysozyme gene would be tethered to centromeric heterochromatin, at least in a number of cases this centromeric heterochromatin should include the centromere of its own chromosome. High precision 3D-distance measurements [ 35 , 36 ] from the lysozyme gene domain to the corresponding chromosome 1 centromere in the data sets described above showed that there is no such colocalization (Table 1 ). In those cell types where the lysozyme gene is completely shut off, the smallest distances found were 0.6 μm in proerythroblasts and 0.5 μm in multipotent precursor cells. This finding rules out a colocalization of the two loci. Distances in multipotent precursors are on average somewhat smaller than in the other cell types (Table 1 ). This can be attributed to a more compact chromosomal shape and to a smaller nuclear volume in this cell type (see below). Table 1 3D-Distance measurements between the lysozyme gene domain and the centromere of the corresponding chromosome 1 in interphase nuclei of different cell lines. multipotent precursor cells proerythroblasts myeloblasts macrophages LPS induced macrophages evaluated territories 70 85 80 89 78 mean value 1,5 μm 2,1 μm 2,5 μm 2,2 μm 2,2 μm median 1,4 μm 2,0 μm 2,2 μm 2,2 μm 2,1 μm Standard-deviation 0,8 0,8 1,1 0,8 0,9 Smallest value 0,5 μm 0,6 μm 0,7 μm 0,7 μm 0,5 μm Largest value 4,8 μm 5,8 μm 5,3 μm 4,4 μm 5,1 μm Radial positioning of chromosome territories 1 and 8 within the nucleus Habermann et al. [ 37 ] showed that in embryonic chicken neuronal and fibroblast nuclei the gene poor macrochromosomes 1–5 are located at the nuclear periphery. Intermediate chromosomes 6–10 were found further inside but not as central as the gene rich microchromosomes. Respective results were also found in human and other primate cells [ 38 - 43 ]. According to the current release of the chicken genome sequence [ 31 ] chromosome 1 has a length of 188 Mbp with ~11 genes/Mbp and chromosome 8 has 30 Mbp with ~19 genes/Mbp. These numbers are likely to change somewhat with further releases of the sequence. They do suggest however that the relative gene content is higher for the smaller chromosome 8. To test for a difference in the radial distribution of individual chicken chromosome territories, we measured 3D radial distributions in the nuclei painted with chromosomes 1 and 8 from experiments described above (Figure 7 ). Figure 7 Radial distribution of chromosomes 1 (red) and 8 (green) in nuclei of (a) multipotent precursor cells (n = 37), (b) myeloblasts (n = 27), (c) activated macrophages (n = 23) and (d) proerythroblasts (n = 40). Unlike in the median distribution used for determination of significance levels, in the graphs shown here all voxels of a segmented signal are represented. Chromosome 8 has only about one sixth of the size of chromosome 1. Accordingly, its interphase territories are much smaller, leading to a smaller sample of voxels and thus accounting for less smooth curves than for chromosome 1 territories, e.g. in myeloblasts. All curves for each chromosome are shown in one graph in a supplemental figure in additional file 1. In multipotent myeloid precursor cell nuclei, chromosome 1 territories were located more peripheral than chromosome 8 territories (p < 0.005). The same was true for proerythroblasts (p < 0.001) but no significant difference was present in myeloblasts (p > 0.1). These three cell types grow in suspension and have round nuclei. In flat nuclei of LPS-stimulated macrophages we again found chromosome 1 territories more peripheral than chromosome 8 territories (p < 0.005). The radial distribution is also reflected by the signal median values. It indicates at which nuclear radius half of the signal voxels are more internal and half are more external. In 73% of the precursor cells the chromosome 1 signal median is more external than the chromosome 8 signal median. The respective values are 52% for myeloblasts, 82% for proerythroblasts and 91% for activated macrophages. When comparisons between cell types were made, chromosome territories 8 showed a rather similar radial distribution in all cell types (p > 0.1 or >0.05 for all combinations with a frequency maximum of ~10% at or near 80% of the nuclear radius, Figure 7 , supplemental figure in additional file 1 ). Chromosome 1 territory distribution did show significant differences between cell types (Figure 7 , supplemental figure). Chromosome 1 territory radial distribution was compared between nuclei co-hybridized either with the cLys -probe (series 1, Figure 2 , not shown as graph) or the chromosome 8 paint probe (series 2, Figure 3 and Figure 7 ). In LPS-stimulated macrophages chromosome 1 territories were further outside than in proerythroblasts (p < 0.001 in series 1 and series 2), in myeloblasts (p < 0.001 in series 1 and p < 0.01 in series 2) and in precursors (p < 0.01 in series 1 and p < 0.05 in series 2). No significant difference was found between stimulated and unstimulated macrophages (p > 0.1, series 1 only). A significant difference in radial distribution of chromosome 1 territories between precursors and myeloblasts was found in series 1 (p < 0.01) but not in series 2 (p > 0.1). The same was true for the comparison of precursors and proerythroblasts (series 1: p < 0.001; series 2: p > 0.1). Proerythroblasts and myeloblasts did not show a significant difference (p > 0.1 in both series). The radial distribution of the lysozyme gene domain did not differ significantly between the cell lines (p > 0.1). The mean value of the signal medians was between 71 and 76% of the nuclear radius for all cell lines. In erythroblasts, chromosome 1 territory signal medians were more internal than cLys signal medians (66% vs. 72%, p < 0.005). In unstimulated macrophages the opposite was true (76% vs. 73%, p < 0.05). In the other cell types the differences between the signal medians of cLys and chromosome 1 territories were between 0–2% and not significant. Our results are compatible with previous data [ 37 ] in describing an external location for chromosome 1 and a somewhat more internal location for chromosome 8. In addition we find differences in the radial distribution of chromosome territories between the chicken cell types that have not been observed previously. Nuclear volumes The volume of nuclei was measured in confocal stacks of DNA counterstain of data sets described above by the same program that was used for the calculation of the radial distributions (Figure 4 ). The mean value for nuclear volumes for multipotent myeloid precursors was 212 μm 3 (± 75 standard deviation, n = 76) and for myeloblasts 327 ± 101 μm 3 (n = 50). The difference between the two cell types was highly significant (p < 0.001). The mean nuclear volume of proerythroblast was 296 ± 109 μm 3 (n = 78). Proerythroblasts did not consistently show significant differences when compared to precursor cells or myeloblasts. Since the macrophage cell line carries additional chromosomes, its nuclear volume cannot be directly compared to the other cell lines. For unstimulated macrophages we determined nuclear volumes of 459 ± 91 μm 3 (n = 42) and for LPS stimulated macrophages 554 ± 139 μm 3 (n = 78). This difference was not significant. The measured nuclear volume depends on the chosen signal threshold. Since the volume increases by the power of 3 with the nuclear radius, small differences in the segmentation can lead to large volume differences. A cautious interpretation of such measurements is thus advised. The difference between multipotent myeloid precursors and myeloblasts is so large however that we are confident that it is real and not a thresholding artifact. Discussion Cell type specific chromatin distributions on the nuclear level have been described for over a century [[ 1 ], p.100]. Differences between cell types have also been described for the distribution of heterochromatin detected with antibodies against methylated histones [ 44 ], for the radial distribution of gene rich and gene poor chromosomes [[ 37 , 40 , 45 ] this study] and the occurrence of clustering between specific chromosome territories [ 45 ]. Here we show an example were large-scale chromatin organization of chromosome territories changes during differentiation, and thus add a new feature to the list of nuclear architectural properties that can differ between cell types. To quantify chromatin dispersal of labeled chromosomes in cells of various differentiation stages, we counted the number of separate, labeled chromatin objects to which the chromosome territories disintegrated at increasing thresholds. In the investigated chicken cell types, chromosome territories of further differentiated cell types disaggregated into more objects. An increase in object number during differentiation may indicate that a significant number of compact chromatin domains with silent genes separate from each other into several, more decondensed, "open" chromatin domains. This would increase the accessibility to transcription factor complexes from the interchromatin compartment by increasing the available chromatin surface area of the chromosome territory. In human lymphoblasts gene rich chromosome 19 territories were found more decondensed than gene poor chromosome 18 territories [ 38 ] and electron microscopic evidence suggests that active genes are exposed at chromatin domain surfaces in a zone called the perichromatin region, a transitional zone that marks the transition form the chromatin domain periphery to the interchromatin compartment [ 46 ]. A caveat of this interpretation is that so far no unequivocal proof for a profound influence of higher order chromatin compaction on gene activation and gene silencing has been presented. A further possibility is that inactive loci in the more differentiated cells do not require a tight spatial silencing by chromatin compaction anymore because the set of available molecular activators and repressors has changed. At present we can only speculate whether the correlation of increased dispersal of chromosome territories with differentiation state is a widespread feature or restricted to a few chicken blood cell types. At the highest, nuclear level of chromatin organization it was described for mammalian nuclei that heterochromatin shows distinct patterns of large blocks in terminally differentiated cells but not in blood stem cells and tumor cells [ 47 , 48 ]. This indicates a compaction of chromatin in differentiated cells rather than in their precursors, unlike in our current data on the chromosome territory level. It is possible, that heterochromatin (consisting mainly of repetitive sequences) and the bulk of labeled chromosome territories behave differently in these aspects. Due to their suppression with unlabeled repetitive DNA, repetitive sequences are underrepresented in chromosome territories detected by FISH as in the present study. Also, the rather small amount of repetitive sequences and heterochromatin in the chicken genome (genome size ~1.2 Gbp according to [ 29 ], 1.1 Gbp according to [ 31 ]) as compared to mouse and human genomes (~3.2 Gbp each, [ 31 ]) may lead to differences in nuclear organization. Multipotent myeloid precursor cells have the smallest nuclei of the cell types investigated here. Myeloblasts have on average larger nuclei than proerythroblasts. If the observed disaggregation of chromosome territories were based on a nuclear volume increase, the larger myeloblast nuclei should have a stronger dispersion of chromosome territories than proerythroblasts. However, the opposite is true (Figure 5 ). We thus conclude that chromosomal dispersion is not related to nuclear size. In general, we observed larger nuclear volumes for further differentiated cell types. Increasing nuclear size was also observed during maturation of nerve ganglia cells [ 49 ] while a volume decrease was described during the maturation of lymphocytes [ 47 ]. Accordingly, unlike recently suggested [ 50 ], a decrease of nuclear size does not appear to be a phenomenon generally associated with terminal differentiation events. The lysozyme gene domain is positioned inside the chromosome 1 territory in multipotent myeloid precursor cells where the lysozyme gene is inactive, but on the surface or outside in most of the territories in activated macrophages where the gene is strongly expressed. We thus did find a tendency to more exterior regions of the chromosome territory for the highly expressed gene from in activated macrophages although actual looping-out (without visible contact to the territory) was observed in only about 6%. Interestingly, while the radial distribution of the lysozyme gene domain within the nucleus is about the same in all cell types, the harboring chromosome 1 territories show differences. The finding that in erythroblasts the cLys signal is more exterior than the chromosome 1 territory signal median but in unstimulated macrophages the opposite is true also argues for a cell type specific organization of chromosome territories. A similar observation has been described for a IL-3 induced differentiation of human leukemic K562 cells where the β-globin gene cluster does not change nuclear position but the harboring chromosome 11 territory does [ 51 ]. However, in human hematopoietic cells a relocation of a gene to a different preferential radial position [ 52 ] or to or away from heterochromatic nuclear compartments has been observed for some genes, correlated with transcriptional regulation at different developmental stages [e.g. [ 33 , 53 ]]. Unfortunately, the harboring chromosome territories were not labeled in these studies. While we can exclude a tethering of the inactive lysozyme gene to the centromere, at first glance this result seems compatible with the hypothesis that inactive genes are stored away in internal regions of the chromosome territory and active genes are on their surface or even looped out. However, several aspects suggest an alternative explanation. (i) Embedded in the chicken lysozyme gene domain is a second gene, cGas41 , which, albeit on a low level, is expressed in all cell types used in this study, including multipotent myeloid precursor cells [ 28 ]. Thus we found an example of an active gene with a location inside the territory as it was described previously for some mammalian genes [ 13 ]. (ii) Although the position of the lysozyme gene domain is most peripheral in activated macrophages where the expression is highest, we also found a shift towards more external positions from multipotent myeloid precursor cells to further differentiated proerythroblasts, both non-expressing cell types. (iii) In addition to the lysozyme gene domain, we investigated the chromosome 1 centromere. Surprisingly, both loci showed a very similar distribution in all cell types investigated. Transcription from centromeres has been observed in yeast [reviewed in [ 54 ]] and from a human neocentromere [ 55 ]. Formally, we thus cannot fully exclude that centromeric transcription may occur in chicken. We regard it as extremely unlikely, however, that tissue dependent differences in centromeric transcription play a role in the cell type specific spatial positioning found here. The observed modification in the morphology of chromosome territories during differentiation rather invites to hypothesize that the positional changes observed for the lysozyme gene domain are not restricted to this particular chromatin loop or only to those chromatin loops which harbor genes that become activated during cell differentiation. Instead, these positional changes may reflect a more general, differentiation dependent change in large-scale chromatin structure. Differentiation processes may thus have a more global impact on chromatin structure than previously suspected. Conclusions We describe several features of chromosome territory organization that differ between various hematopoietic chicken cell lines. While multipotent myeloid precursor cells had compact chromosome territories, the more differentiated cell types investigated here displayed somewhat disaggregated, diffuse territories. Although nuclear volumes generally are larger in the more differentiated cell types, they do not correlate with the changes in chromosome territory morphology. The chicken lysozyme locus as well as the chromosome 1 centromere is located preferentially in the interior of the chromosome territory in precursor cells and more external in more differentiated cells. Our data suggest that such a repositioning of chromosomal loci during differentiation may be a consequence of general changes in chromosome territory morphology, not necessarily related to transcriptional changes. The radial distribution of chromosomes 1 and 8 also differed between cell types. In summary our data argue for a cell type specific chromosome territory organization in the investigated cell lines. Methods Cells All cells are from retrovirally transformed chicken cell lines [ 56 , 57 ]. HD50MEPs represent multipotent myeloid precursor cells before the separation of erythroid and monoblast/macrophage lineages. Proerythroblast-like HD37 were used as an example for a differentiated cell type that is negative for lysozyme expression. Myeloblasts (HD50 myl, a cell line derived from the same precursors as HD50MEP) are an intermediate stage in the differentiation to macrophages (HD11). While macrophages grow adherent, all others grow in suspension. Cytogenetic analysis showed that all lines were diploid for chromosome 1 except HD11. The latter had 2 to 4 normal chromosomes 1 plus a translocation chromosome with the short arm of #1 including cLys but not the #1 centromere and also other karyotype aberrations, e.g. a trisomy of chromosome 2. All lines were diploid for chromosome 8. Cells were cultured in IMDM supplemented with 2% chicken serum, 8% fetal calf serum and 2 mmol L-Glutamin and maintained at 37°C with 5% CO 2 . For 3D-FISH experiments cells were cultured on #1.5 glass coverslips (170 μm thick). Except for macrophages, coverslips were pretreated for 35 min with poly-L-lysine (0.1 mg/ml; MW 300000, Sigma, Deisenhofen, Germany, P5899). Activation of HD11 was achieved by the addition of 5 μg LPS (Sigma, L-8275) per ml medium and subsequent cultivation over night [ 58 ]. Activated macrophages become postmitotic. Coverslips with attached cells were washed with PBS, fixed in 1,2% formaldehyde freshly made from paraformaldehyde [ 59 ] for 15 min, washed in PBS 3 × 3 min, incubated in 0,5% Triton X-100 in PBS for 20 min and equilibrated in 20% glycerol in PBS for 60 min. Dipping in liquid nitrogen and thawing at room temperature in 20% glycerol/PBS was performed five times. After washing 3 × 3 min in PBS and incubation in 0,1 M HCl for 10 min, coverslips were washed 2 × 3 min in 2 × SSC and stored in 50% formamid/ 2 × SSC for at least 1 h but usually overnight or longer at 4°C. FISH Chromosome-specific paint probes for chicken chromosomes 1 and 8 were kindly provided by Dr. Felix Habermann and Dr. Johannes Wienberg, Munich. They were generated by flow sorting of metaphase chromosomes and subsequent degenerated oligonucleotide primed (DOP)-PCR [ 60 ]. They produce a uniform labeling of metaphase chromosomes. When the amplification products were repeatedly reamplified using the same PCR conditions, however, we noted after about a dozen rounds that the label on metaphase chromosomes became non-uniform in appearance indicating a reduction in complexity by the loss of sequences. The higher likelihood of retaining repetitive sequences was reflected by the finding that the by far brightest spot was found at the centromere as identified by the primary constriction in the DNA counterstain. To identify chromosome 1 centromeres together with completely delineated chromosome 1 territories, we used a mixture of an early amplification product with repeatedly reamplified probe. We thus obtained intense painting of the entire chromosome and a particular bright signal at the centromere (Figure 2a ). In experiments where chromosomes 1 and 8 were cohybridized, detection of centromeres was not necessary and thus only early amplification products were used. DOP-PCR for amplification and labeling (biotin-16-dUTP for #1 and digoxigenin-16-dUTP for #8, both from Roche Applied Science, Mannheim, Germany) was performed as described [ 61 ]. The lysozyme gene domain is contained on the pPoly-III-i Lys-plasmid [ 62 ]. It was labeled by nick-translation with digoxigenin-dUTP. Chicken cot 1 DNA was prepared from liver using standard procedures. The same result was obtained for the chicken chromosome 8 paint probe (data not shown). All DNA for a given assay was mixed, precipitated and solved in deionized formamide. The same volume of 20% dextransulfat in 2 × SSC was added. In experiments with cLys the following amounts of DNA were precipitated for each μl hybridization mix: 2 μl label-DOP-PCR product of an early amplification round of the #1 paint plus 2 μl of a highly amplified paint probe for centromere detection, 50 ng pPoly-III-i Lys, 2.5 μg cot 1 DNA. When paint probes for #1 and #8 were cohybridized, 2 μl label-DOP-PCR product of an early amplification round was used for each chromosome and supplemented with cot1 DNA as above. Denaturation was 5 min at 85°C. Preannealing with the cot 1 DNA was performed for 25 min at 37°C. For 3D FISH, coverslips with cells were denatured in 70% formamide for 3 min at 70°C, and placed immediately in ice-cold 50% formamide/2 × SSC. They were then incubated with 5 μl hybridization mix under a sealed 18 × 18 mm 2 coverslip at 37°C for 24 h-72 h. Air-drying was carefully avoided at all steps. Metaphase chromosomes were hybridized as described [ 37 ]. Detection was performed as described [ 63 ], using rabbit anti-digoxigenin (1:500) and goat anti-rabbit-Alexa488 (both from Sigma) and for biotin detection Avidin-Cy3 (Dianova, Hamburg, Germany). Slides were counterstained with DAPI and TOPRO-3 (Molecular Probes, Eugene, Oregon) and mounted in Vectashield (Vector Laboratories, Burlingame, CA). Confocal laser scanning microscopy 3D image stacks (8 bit) were recorded with a Leica TCS SP confocal laser scanning microscope equipped with an argon (488, 514 nm) and a HeNe laser (633 nm) (Leica Mikrosysteme, Bensheim, Germany). A 100 × N.A. 1,4 oil objective was used to obtain stacks with a voxel size of 0.08 × 0.08 × 0.24 μm. Nuclei with separated homologous chromosome territories were preferably selected for recording. To measure the chromatic aberration, 0.5 μm multi-color latex beads (Polysciences Europe, Eppelheim, Germany) were fed to activated macrophages. After phagocytosis the cells were fixed and embedded like 3D-FISH preparations. The beads were in the cytoplasma and thus their optical environment was closer to the situation of FISH signals than beads mounted directly on a glass cover slip. The chromatic aberration in x, y and z was corrected before the assignment of signals to categories or distance measurements were performed. Image analysis The program used for object counting was first applied by Cremer et al. [ 44 ]. The original 8-bit gray level image stack is first subjected to Gaussian filtering and then normalized, i.e. the lowest existing gray value is set to zero, the highest to 255 and the values in-between are recomputed accordingly. A starting threshold of 20 was chosen and voxels above the threshold were determined. Of those, all touching voxels (26 voxel neighborhood) were combined to objects. Only structures with at least 10 voxels were regarded as 'objects' and included in the further analysis. After counting the objects and calculating the other parameters, the threshold was raised for 5 gray levels, object determination and calculation was repeated and so on until the highest applied threshold of 250 was reached. For statistical calculations, from each nucleus the maximum number of objects occurring at any threshold of 80 or higher was used. The restriction to thresholds of 80 or higher was made to confidently exclude background objects. Statistical inferences about the means of the maximum values of objects were based on one-way analyses of variance (ANOVA). Post-hoc comparisons generating p-values relied on Sidak tests. These test were performed with SPSS version 12 (SPSS Inc., Chicago, IL). For the second parameter, for each nucleus at each threshold the ratio (object intensity)/(object surface voxels) was measured for all objects and averaged. Object surface voxels are defined as voxels belonging to an object and having at least one of the 26 neighbors not belonging to the object. The unit is 1/μm 2 . This value reflects the amount of intensity (chromatin) that is enclosed in a given surface area. Chromosome territories with a richly folded surface thus have a rather low value whereas compact, homogeneous territories have a higher value. Most nuclei had zero or few objects at very high threshold levels (Figure 5a,5b ). Therefore, the calculation of meanvalues for the intensity/surface parameter was stopped when less than five nuclei with at least one object where left. For the third parameter, average amount of labeled chromatin per volume, the intensity of all voxels belonging to an object was summed up and divided by the number of object voxels and the average over the objects was computed. Localization of cLys and centromere signals with regard to chromosome 1 territories: Image stacks were imported in ImageJ (freely available on the internet at [ 64 ]) and each fluorochrome was assigned to one channel of an RGB-Stack. A Gaussian Blur filter was applied before using the 'brightness & contrast' function to enhance signals and decrease background. The 'make montage' function was then used to show all planes of the RGB-stack side by side. For each gene or centromere signal the z-plane was selected where it was brightest. This plane was then used for the categorization (Figure 6a ) [ 11 ]. The Mann-Whitney-U test from SPSS was used for statistical analysis. For the aneuploid cell line HD11 an evaluation was performed only if normal chromosomes were unequivocally distinguishable from the translocated p-arm (without centromere). The latter was excluded from further analysis. High precision 3D-distance measurements [ 35 , 36 ]: Gravity centers of the signals were determined with Showpos, a program written by Kurt Sätzler, Heidelberg, for Silicon Graphics Workstations running under Irix. The 3D coordinates of cLys and the respective centromere were corrected for the chromatic aberration and the distance was calculated. The quantitative assessment of 3D radial distributions of painted chromosome territories and the measurement of nuclear volumes was performed using a program developed by Dr. Johann von Hase, Heidelberg which is described in detail elsewhere [ 42 ]. To determine the statistical significance of radial distribution differences, we used the medians of each signal in each nucleus and applied the two-sided Kolmogorov-Smirnov test [ 65 ]. The same test was applied to nuclear volumes. Authors contributions The study was designed by SD together with CB and TC. First experiments and development of techniques were performed by RM. S Stadler and VS performed 3D-FISH, microscopy, determination of cLys and centromere 1 localization and high precision distance measurements of all evaluated cells. RM carried out radial distribution statistics and volume measurements. All three were supervised by SD and TC. S Stein programmed and adapted the object counting program, supervised by CC. Object counting and statistical analysis was performed by SD. The manuscript was written by SD with substantial contributions by CB, TC and VS. All authors read and approved the final manuscript. Supplementary Material Additional File 1 Radial distribution of chromosomes 1 and 8 in nuclei of different cell types. These graphs show the same curves as presented in Figure 7 , but now all curves for chromosome 1 are combined in (a) and those for chromosome 8 are combined in (b). Click here for file	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC535556.xml
524498	Feasibility of rapid and automated importation of 3D echocardiographic left ventricular (LV) geometry into a finite element (FEM) analysis model	Background Finite element method (FEM) analysis for intraoperative modeling of the left ventricle (LV) is presently not possible. Since 3D structural data of the LV is now obtainable using standard transesophageal echocardiography (TEE) devices intraoperatively, the present study describes a method to transfer this data into a commercially available FEM analysis system: ABAQUS © . Methods In this prospective study TomTec LV Analysis TEE © Software was used for semi-automatic endocardial border detection, reconstruction, and volume-rendering of the clinical 3D echocardiographic data. A newly developed software program MVCP FemCoGen © , written in Delphi, reformats the TomTec file structures in five patients for use in ABAQUS and allows visualization of regional deformation of the LV. Results This study demonstrates that a fully automated importation of 3D TEE data into FEM modeling is feasible and can be efficiently accomplished in the operating room. Conclusion For complete intraoperative 3D LV finite element analysis, three input elements are necessary: 1. time-gaited, reality-based structural information, 2. continuous LV pressure and 3. instantaneous tissue elastance. The first of these elements is now available using the methods presented herein.	Background Intraoperative TEE is currently available in most cardiac surgical operating rooms. In some centers, intraoperative 3D echocardiography is used to evaluate geometry and to plan surgical interventions prior to LV remodeling surgery. However, quantitation of LV geometry is limited to rather imprecise measures such as ejection fraction. Thus the cardiac surgeon has no sophisticated, immediate, quantitative analysis of the preoperative 3D LV geometry. Intraoperative quantitative analysis of the dynamic behavior of the LV might provide optimal information upon which to base precise patient-specific planning of the surgical intervention, as well as to assess the adequacy of the completed surgical repair. Because the LV cannot be realistically described by a symmetric mathematical model, the modern approach consists of using a FEM mesh which approximates LV geometry [ 1 ] or whole heart geometry [ 2 ]. Initial attempts at FEM in the heart have been carried out with 3D segmentation and tracking using sophisticated and expensive cardiac MRI [ 3 ]. MRI is impractical in the cardiac surgical operating room and is complicated by the fact that the LV and the papillary muscles are active materials, behaving differently during systole and diastole. An ideal model would provide material properties specific to each patient as first mentioned by McCulloch [ 4 ], but untill now patient-specific modeling in the operating room is not been possible. FEM modeling of 3D intraoperative echo data provides an excellent tool for incorporating material properties, volumetric data and boundary pressures to more accurately record and then to simulate LV dynamic performance. Accurate simulation will be the foundation of surgical planning. The limitation until now in applying FEM intraoperatively has been the technical complexity of this technique. The purpose of this study is to take the first step towards introducing FEM into the operating room environment. The goal is to facilitate transfer of geometric data from 3D ultrasound data set into FEM. Methods After obtaining institutional review board approval, LV images from clinical TEE data sets were obtained in five patients via the midesophageal window using a Philips 5500/7500 or and Acuson Sequoia ultrasound system. After induction of general anesthesia and airway protection, the esophagus was intubated using an omniplane TEE probe. 3D TEE data sets of the LV structures including mitral annulus and leaflets, chordae tendinae, papillary muscles and ventricular wall were obtained using the automated Philips/Acuson acquisition protocol at 10° increment. Images were gaited for both beat-to-beat variability and respiratory motion. In order to facilitate acquisition in the shortest possible timeframe, ventilation was modified to provide a tidal volume of 510 -6 m -3 kg -1 at a respiratory rate sufficient to maintain end-tidal CO 2 levels between ~4.410 3 m kg -1 s -1 and ~5.1*10 3 m kg -1 s -1 . All images were stripped of patient identifiers. For LV geometry reconstruction, the TomTec LV-Analysis TEE © software module [ 5 ] was employed. This software runs on a standard Dell Inspirion laptop computer with Microsoft Windows™ 2000 operating system which imports, analyzes, reports and archives the time-resolved 3D-ultrasound data. The TomTec system automatically detects endocardial borders and produces a 3D shell reconstruction of the LV [ 5 ]. It also provides for an analysis of global and regional LV parameters in which a landmark-setting method is used (see Fig. 1 ). Figure 1 Scheme of the LV (left ventricle). Section through left atrium and ventricle shown schematically. In the LV Analysis TomTec TEE program, three landmarks are taken from each second frame per data set. This means that each 10° a frame is taken as the sampling point for the LV Analysis TEE program. AV is aortic valve, MV is mitral valve and Ap is apex. The first landmark was set in the middle of the mitral valve at the level of its annulus. Care was taken to avoid having the mitral valve cusps cross this landmark. Two additional landmarks were placed in the middle of the aortic valve at the level of its annulus and at the endocardial level of the LV apex. With this landmarking procedure, a time-resolved LV geometric analysis with 18 models per heart cycle was obtained (see Fig. 2 ). Figure 2 Screenshot-TomTec. Screenshot of the workspace of the TomTec LV Analysis TEE program. The LV is segmented using color coding in (c). In (a) the LV model is shown in 3D as calculated from the sampling points set according to Fig. 1. The shadowed plane in (a) indicates the position of the actual original US gray-value frame in 3d as shown in (b). In (d) the volume content is displayed in terms of the actual model step indicating the actual phase with a green line. The screenshot of the actual phase shows the LV model at near systole. The rendered LV geometry resulting from the TomTec analysis tool was transferred to an ABAQUS input file using software written in Delphi. In this program, the TomTec file structure was reformatted to an ABAQUS system (version 6.3) input file based on standard ABAQUS FEM elements. ABAQUS creates a time series of LV model files and requires continuous intraventricular pressure and tissue elastance parameters to process the model. For this analysis LV pressure was modeled using wave forms obtained from Columbia University's HeartSim © cardiac simulator [ 6 ]. A single tissue elastance parameter was applied. These modeled values were used to demonstrate the concept. Actual values will be needed for accurate simulations. The time required for each step in this process was recorded for each patient data set. Results Both, the Philips Sonos 5500/7500 or the Acuson Sequoia ultrasound systems required less than 10 minutes acquistion time per patient. The application of the TomTec LV analysis algorithms with manual placement of the necessary landmarks took approximately 7 minutes per patient. In five patient data sets conversion from TomTec data to the FEM model was carried out in less than a minute for a heart cycle using the conversion tool MVCP FemCoGen © [ 7 ]. ABAQUS processing time on the above computer was 20 seconds per sequence or approximately 6 minutes per patient. Total time for the procedure was approximately 24 minutes per patient (see Table). Fig. 3 shows the ABAQUS FEM program system interface (ABAQUS/viewer) including the LV models in default mesh mode. All 774 triangles of the FEM mesh from the diastolic state (14 th image out of a set of 18 images per heart cycle) are displayed and can be visualized using ABAQUS viewer options. Each mesh element can be analyzed separately. This is shown in Fig. 4 : In Fig. 4a and 4b the rendered LV using a standard constant-shading model is displayed in systolic and diastolic states. Fig. 4c and 4d show both the FEM mesh and the normal vectors orthogonally placed (orthonormals) on each triangle indicating the force direction. These figures demonstrate the quantification of movement during the heart cycle directly using modeled continuous LV pressure and tissue elastance parameters. Table 1 Data acquisition and processing times Data acquisition and processing times Data set Ultrasound acqusition [s] TomTec analysis [s] FemCoGen transfer [s] Abaqus processing [s] 1 715 475 50 355 2 580 364 45 340 3 670 320 55 390 4 640 390 45 410 5 597 423 45 430 Mean ± (standard deviation) [s] 640.4 ± 41.7 394.4 ± 43.7 48 ± 3.6 385 ± 30 Total time [s] 1467.8 ± 29.7 Figure 3 LV in finite element analysis program. Left ventricle FEM model in ABAQUS FEM program interface. Shown is the LV in diastole. At the top of the mesh is the aortic valve depicted as a cavity. The LV apex appears at the bottom of the mesh. Figure 4 Pressure direction at systole and diastole. Rendered LV at systole on the left (a) and (c) and diastole on the right (b) and (d). Shown is the mesh generated with FEM program including all 774 FEM elements rendered with a standard constant shading model in (a) and (b). (c) and (d) show the mesh together with the surface vectors (normals) orthogonally placed on each element (triangle) indicating the pressure directions. Discussion The general intention of this study was to demonstrate the feasibility of transporting individual patient's LV geometry data into a FEM model. Standard laptop computer technology was utilized to accomplish the transfer from common TEE-machines (Philips Sonos 5500/7500 and Acuson Sequoia). The software running on the laptop was the commercially available TomTec LV Analysis TEE © package and ABAQUS FEM system, plus the recently developed MVCP FemCoGen © . Accomplishing this transfer will form the foundation for intraoperative surgical planning and quantitative outcome assessment of valvular and LV reconstructive surgery. The scope of this study was to produce a prototype in which the feasibility of the method could be assessed. In a fully operational system, we could postulate clinical applications such as enhanced/automated wall motion abnormality detection, assessment of regional relaxation which encompases the entire ventricle, assessment and guidance of ventricular remodeling operations, and serial assessment of recovery of regional wall function post myocardial stunning. FEM meshes have been used for approximately 30 years [ 8 ] in the analysis of many anatomical structures and organs e.g such as major vessels [ 9 , 10 ], heart valves [ 11 ] and ventricles [ 12 ], lung [ 13 ], corneoscleral shell [ 14 ], plastic and reconstructive craniofacial surgery [ 15 ] and the femur [ 16 ]. A FEM model can be created to determine the deformation of the LV loaded by intraventricular pressure. Steady-state fluid dynamics and structural analyses can be carried out using commercial codes based on FEM [ 17 ]. At a sequence of time-steps of the cardiac cycle, the model can be considered to be a quasi-incompressible transversely isotropic hyperelastic material based on the analysis of Feng [ 18 ]. Until now, biomechanical cardiac FEM models have been based on simplified ellipsoidal and cylindrical geometries [ 18 ]. A FEM created in this way is not patient-specific and does not accurately represent precise regional deformations in the LV loaded by intraventricular pressure. The method described here will allow patient specifity and the precise representation of deformation. Our method would be applicable to the "live 3D" systems assuming that the entire ventricle could be seen throughout the cardiac cycle in the transthoracic (or epicardial) matrix array acquisition. This would be most feasible in small adults and children and can be proved in further studies. The total time required for acquisition to a completed FEM model was approximately 24 minutes and can be accomplished during the time period when the patient is being prepared for cardiopulmonary bypass (generally 1 to 1.5 h). Thus the feasilbility in terms of duration is clearly demonstrated. In terms of procedure accuracy, reproducibility and duration, the primary limitation is the dependence of the TomTec software on manual entries of the three registration landmarks. This requirement is iterative. Manual entries must be done for multiple frames within the TEE data sets. Inter- and intraobserver variability is a general problem for ultrasonic imaging. The validation of the TomTec border detection has not been published. TomTec LV Analysis TEE © Software is under review by the US FDA, but despite of lack of validation TEE is the only practical technology in the cardiac operating room for the forseeable future. Ultrasound tissue Doppler technologies may be developed in the future to allow automation of the registration process. A limitation of the present study is that it is focused on the deployment of the transfer method. The entire process will require extensive validation. The validation strategy will most likely involve comparision with preoperative cardiac MRI as well as comparison with bypass and post bypass tissue geometry in the same patients. Creating models from MRI based data sets analogous to the TomTec LV analysis and transfering these models to ABAQUS might lead to a new validation strategy which is not been possible up to now. The tool for modeling presented here facilitates vector-subtraction analysis for different points within the cardiac cycle. Quantification is therefore immediately available for both global and regional wall motion, shape and volume analysis. The future use of such instantaneous analysis has a number of potential applications for LV function assessment and surgical planning. This technology could enable a comprehensive automated regional wall motion analysis. A significant challenge in the evaluation and management of patients with coronary artery disease is determining the viability of myocardium. A biomechanical FEM of the LV myocardium can be imported to evaluate dynamic mechanical properties of regions of the myocardium. This approach could provide the basis for a new index of regional myocardial viability. Conclusions For complete intraoperative 3D LV finite element analysis, three input elements are necessary: 1. time-gaited, reality-based structural information, 2. continuous LV pressure and 3. instantaneous tissue elastance. The first of these elements is now available using the methods presented herein. The later two parameters will be required for robust modeling and analysis. Pressure data will be easily available in the cardiac operating room. Strategies for computing elastance are presently under development. With all three parameters, it will be possible to begin to develop the computational strategies which will allow virtual procedures to be performed utilizing 3D display technology and a haptic-feedback robotic "instruments". Whether this new intraoperative information will be useful in assessing the effectiveness of surgical interventions such as LV remodeling remains to be studied. FEM analysis has not been feasible for LV in the intraoperative setting. The major roadblock was the complexity and the practicality of transfer of structural 3D data to a FEM analysis program. This study describes a method to rapidly transfer 3D structural data from the TEE device into a FEM analysis program. Once mesured pressure and calculated elastance are added to the model, near real-time dynamic stress-strain information in the operating room will be achievable. Authors' contributions JFV did the technical part implementing the FEM model in ABAQUS ® , NSN did the data acquisition and the medical part. Both authors read and approved the final manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC524498.xml
509293	The Regenerative Plasticity of Isolated Urodele Myofibers and Its Dependence on Msx1	The conversion of multinucleate postmitotic muscle fibers to dividing mononucleate progeny cells (cellularisation) occurs during limb regeneration in salamanders, but the cellular events and molecular regulation underlying this remarkable process are not understood. The homeobox gene Msx1 has been studied as an antagonist of muscle differentiation, and its expression in cultured mouse myotubes induces about 5% of the cells to undergo cellularisation and viable fragmentation, but its relevance for the endogenous programme of salamander regeneration is unknown. We dissociated muscle fibers from the limb of larval salamanders and plated them in culture. Most of the fibers were activated by dissociation to mobilise their nuclei and undergo cellularisation or breakage into viable multinucleate fragments. This was followed by microinjection of a lineage tracer into single fibers and analysis of the labelled progeny cells, as well as by time-lapse microscopy. The fibers showing morphological plasticity selectively expressed Msx1 mRNA and protein. The uptake of morpholino antisense oligonucleotides directed to Msx1 led to a specific decrease in expression of Msx1 protein in myonuclei and marked inhibition of cellularisation and fragmentation. Myofibers of the salamander respond to dissociation by activation of an endogenous programme of cellularisation and fragmentation. Lineage tracing demonstrates that cycling mononucleate progeny cells are derived from a single myofiber. The induction of Msx1 expression is required to activate this programme. Our understanding of the regulation of plasticity in postmitotic salamander cells should inform strategies to promote regeneration in other contexts.	Introduction There is currently a significant focus on strategies to promote regeneration in adult mammals and therefore a renewed interest in the mechanisms that underlie regeneration in urodele amphibians. An adult salamander such as the newt or axolotl can regenerate its limbs and tail, jaws, ocular tissues such as the lens, and small sections of the heart ( Goss 1969 ; Eguchi et al. 1974 ; Oberpriller and Oberpriller 1974 ; Okada 1991 ; Ghosh et al. 1994 ; Brockes 1997 ; Nye et al. 2003 ). A key feature of urodele regeneration is the local plasticity of differentiated cells at the site of tissue injury or removal ( Brockes and Kumar 2002 ; Odelberg 2002 ; Del Rio-Tsonis and Tsonis 2003 ; Tanaka 2003 ). This has been investigated for pigment epithelial cells of the iris ( Eguchi et al. 1974 ; Simon and Brockes 2002 ; Imokawa and Brockes 2003 ; Imokawa et al. 2004 ), cardiomyocytes ( Oberpriller et al. 1995 ; Bettencourt-Dias et al. 2003 ), and skeletal myofibers and myotubes ( Hay 1959 ; Lo et al. 1993 ; Tanaka et al. 1997 , 1999 ; Kumar et al. 2000 ; Echeverri et al. 2001 ), all of which reenter the cell cycle during regeneration, in contrast to their mammalian counterparts. A second aspect of plasticity is the ability of multinucleate skeletal muscle cells to fragment into viable mononucleate cells that then contribute to the regenerate. This process, sometimes referred to as cellularisation, was described in classical studies of limb regeneration ( Thornton 1938 ; Hay 1959 ), but was first analysed with marked cells by implantation of cultured newt myotubes labelled by microinjection with a lineage tracer ( Lo et al. 1993 ) or by an integrated retrovirus ( Kumar et al. 2000 ). The myotubes were effectively converted to mononucleate cells that proliferated in the blastema, and this process occurred in cells that were blocked from cell cycle reentry ( Velloso et al. 2000 ), thus showing that the two aspects of plasticity are not linked mechanistically. In an important recent contribution, myofibers were labelled in situ by microinjection in the tail of the larval axolotl ( Echeverri et al. 2001 ). After amputation of the tail, the myofibers fragmented into viable mononucleate cells, thus establishing that cellularisation occurs during regeneration and contributes to the proliferative zone or blastema. Our understanding of this intriguing process has received considerable impetus from the recognition of two manipulations that induce mammalian myotubes to undergo fragmentation. The first is exposure to myoseverin, a trisubstituted purine derivative isolated from a combinatorial library ( Rosania et al. 2000 ). It evokes depolymerisation of microtubules, apparently by interacting directly with tubulin, as well as inducing changes in the expression of genes that are implicated in tissue remodelling and wound healing. The second is the conditional expression of the homeobox gene Msx1 in mouse myotubes ( Odelberg et al. 2000 ). Msx1 has been studied as an antagonist of myogenic and osteogenic differentiation (reviewed in Bendall and Abate-Shen 2000 ) and is expressed in the migrating precursor cells of limb muscle during chick development, apparently to prevent precocious differentiation ( Bendall et al. 1999 ). The expression in mouse C2C12 myotubes evokes two aspects of plasticity that occur in 5%–10% of the cells; the first is cleavage of the cells into smaller multinucleated myotubes, which remain viable, and the other is the formation of mononucleate cells capable of division ( Odelberg et al. 2000 ). In the latter case, the clonal progeny of a single myotube were shown to be capable of several pathways of mesenchymal differentiation. The studies on cellularisation by myoseverin and Msx1 have underlined that mammalian as well as urodele cells are capable of this response ( Rosania et al. 2000 ; Odelberg 2002 ). Msx1 is expressed during urodele limb regeneration ( Carlson et al. 1998 ; Koshiba et al. 1998 ), as well as during fin ( Akimenko et al. 1995 ; Nechiporuk and Keating 2002 ) and heart regeneration in the zebrafish ( Raya et al. 2003 ), along with other Msx family genes. Therefore, it becomes important to investigate whether it controls cellularisation during regeneration. Although prior studies of this process have used the multinucleate myotube as the target cell in culture, the critical target during epimorphic regeneration is the more differentiated striated myofiber. The regeneration of muscle fibers in vertebrates proceeds by the mobilisation of reserve satellite cells ( Chargé and Rudnicki 2004 ), and these have been described in myofibers of larval salamander limbs ( Popiela 1976 ). Their participation in cellularisation was excluded in the earlier experiments on urodele cells by selective injection of a lineage tracer into myotubes in culture ( Lo et al. 1993 ) or myofibers in the salamander tail ( Echeverri et al. 2001 ). In order to address these various questions, we have established a culture system in which striated myofibers are dissociated from the limb of larval salamanders and attach to a culture substrate where they can be observed by time-lapse microscopy. The fibers are found to be activated by dissociation to undergo cellularisation and viable fragmentation, and this depends on expression of Msx1 . Results Dissociated Myofibers in Culture In order to obtain striated myofibers, tissue was isolated from the limbs of two species of larval salamander ( Ambystoma maculatum or Ambystoma mexicanum ), which have been used interchangeably with comparable results. After removing the epidermis, the tissue was dissociated by proteolytic digestion, filtered through a sieve to remove most of the mononucleate cells, and plated in serum-free medium. The striated myofibers readily attached to the culture dish ( Figure 1 ; Figure S1 ) and were found to express myosin heavy chain (MHC) and titin after antibody staining. In view of the potential contribution of satellite type cells to the issues under investigation, cultures were treated with a viable nuclear stain and the myofibers were examined carefully at 2 d after plating. Of 1,290 fibers examined in seven independent cultures, there were only 46 examples of mononucleate cells adherent to myofibers, and such cells were not observed in the cases of plasticity that are discussed here. Figure 1 A Live Striated Myofiber from the Larval Salamander Photomicrograph of a live striated myofiber dissociated from the larval limb musculature and adhering to the culture dish in serum-free medium. This cell has the appearance of a normal quiescent fiber and was photographed with VAREL optics at 48 h after plating. Scale bar, 50 μm. When cultures were labelled with tritiated thymidine, no labelled nuclei were observed in multinucleate cells after labelling for 24 h (540 myofibers, five cultures) or 48 h (263 myofibers, three cultures), while 16% of the mononucleate cells were labelled in the latter case. It is noteworthy that the absence of S-phase entry in nuclei within multinucleate cells includes the population of myofibers that undergoes the events of cellularisation or fragmentation described below. Cellularisation of Myofibers after Implantation The dissociation of viable myofibers has allowed us to evaluate their plasticity after implantation into a limb blastema, a procedure that has previously been performed only on myotubes formed in cell culture. Fibers were labelled with a cell tracker dye in suspension after dissociation, and single fibers were examined to verify the absence of any adherent mononucleate cells and were drawn into a glass micropipette ( Figure 2 A). A few fibers (see Materials and Methods ) were injected from the pipette into the early forelimb regenerate of a larval axolotl. The regenerating limbs were sectioned 2–4 d later, and many examples were observed of mononucleate cells labelled with the tracker dye ( Figure 2 B). Such cells were clearly mononucleates, as determined by analysis of serial sections, and were observed in 17 out of 23 animals implanted with labelled myofibers. We conclude that these cells readily undergo cellularisation in the environment of the limb blastema. Figure 2 Cellularisation of Striated Myofibers after Implantation into a Larval Limb Blastema (A) Schematic diagram of procedure. After dissociation of larval limb musculature, the cells were loaded with a cell tracker dye and single myofibers taken up into a suction micropipette, prior to injection into a larval limb blastema as detailed in the Materials and Methods . (B) Section of a limb at 48 h after implantation of CellTracker Orange-labelled myofibers. The section has been counterstained with the nuclear stain Sytox green. Note the dye-labelled mononucleate cells (arrowed). Scale bar, 20 μm. Plasticity in Culture After 48 h in culture, some of the striated fibers remained viable, but showed no significant change in morphology and retained the appearance of the cell shown in Figure 1 . The remainder of the fibers showed various changes in morphology, and these were investigated either by microinjection of single fibers with the lineage tracer Texas red (TR)–dextran and subsequent analysis of the progeny cells or by sequential digital time-lapse observation. Cellularisation. Approximately 10% of the total population of myofibers underwent changes in nuclear localisation within the cells such that a lobulated or ‘cauliflower’ structure formed in the middle or end of the cell ( Figure 3 A and 3 B). This occurred without labelling by tritiated thymidine or any participation by adherent mononucleate cells, which were rarely present on such fibers. The lobules, each of which contained a nucleus, were rapidly resolved into adherent mononucleate cells. In order to analyse these events, single myofibers were microinjected with TR–dextran so as to fill the cells with tracer ( Figure 3 C). We employed the 70 kDa dextran, which is not transferred across gap junctions ( Coelho and Kosher 1991 ; Landesman et al. 2000 ). In cases in which fibers formed the cauliflower structure and underwent cellularisation, the mononucleate cells in the initial colony were labelled with the tracer in a rim of cytoplasm around the nucleus ( Figure 3 D). In some cases, adjacent fibers were injected and the progeny of the myofibers gave rise after 5–7 d to overlapping dense colonies with many labelled cells ( Figure 3 E). These cells did not express detectable levels of MHC after staining by indirect immunofluorescence. Figure 3 Plasticity of Isolated Myofibers (A) Phase-contrast micrograph of a live cell at 3 d after plating, showing a lobulated structure in the middle of the fiber. (B) Micrograph of a live fiber at 2 d after plating, showing budding of nuclei at one end. The cell has been counterstained with Syto 13. (C) Fluorescence micrograph of a myofiber at 24 h after microinjection with TR–dextran. The cell has been counterstained with Syto 13 dye to show the nuclei. (D) Fluorescence micrograph of a colony formed from a single myofiber injected 24 h earlier with TR–dextran. The cell has flattened on the substrate and the nuclei are stained with Syto 13 dye. (E) Fluorescence micrograph of a colony formed from the progeny of several myofibers in proximity that were injected 5 d earlier with TR–dextran. (F) Analysis of the DNA content of cells derived from myofibers injected 5 d earlier with TR–dextran. The DNA content was determined by image analysis of the nuclei of mononucleate TR-positive cells that had been stained with Hoechst (see Materials and Methods ). The green arrow is the value for G 0 nuclei in quiescent myofibers, while the blue arrow is the G 2 /M value determined for mononucleate cells with anti-phosphohistone H3. The red arrow is the G 1 value determined for mononucleate cells. (G) Photomicrograph of a live myofiber, 48 h after plating, showing a binucleate bud formed at the end. The cell was stained as for (B). (H) Fluorescence micrograph of a bud containing three nuclei stained with Syto13 (yellow) derived from a myofiber that contained at least five nuclei and that was injected with TR–dextran (red). Scale bars: (B), (C), and (G), 100 μm; (A), (E), and (H), 50 μm; and (D), 10 μm. We have analysed the DNA content of single Hoechst-stained nuclei by normalised measurements of fluorescence intensity in TR-labelled cells within such colonies, and an example of a representative distribution for a single colony is shown in Figure 3 F. This also shows the corresponding values (shown by arrows in Figure 3 F) for G 0 nuclei in myofibers, G 1 nuclei in mononucleate cells, and G 2 /M nuclei in mononucleate cells labelled with antibody to phosphohistone H3. The histogram of DNA content for cells in the colony is comparable to that previously observed for cycling newt mononucleate cells ( Tanaka et al. 1997 ). The relatively long S-phase in urodele cells leads to a prominent contribution of cells with DNA content between 2N and 4N. In addition, there were examples of TR-labelled mononucleate cells in M-phase, as determined with anti-phosphohistone H3. We conclude that the progeny mononucleate cells are able to traverse S-phase and enter mitosis after cellularisation. Fragmentation. In a second aspect of plasticity, which was shown by 40%–70% of the total population of myofibers, the initial stages also involved the migration of nuclei to form local aggregates, often of two or three nuclei, within the fiber. The migration of nuclei into a terminal aggregate is illustrated by selected images from a time-lapse video analysis ( Figure 4 A; Video S1 ). The series of Figure 4 B illustrates a trinucleate terminal aggregate that fragments from the body of the myofiber (yellow arrows). This fragment remained adherent and extended cytoplasmic processes. In some cases, the nuclear aggregate formed a bud that was discharged into the medium. An example of a multinucleate bud formed at the end of a fiber is shown in Figure 3 G. In cases in which fibers containing at least five nuclei had been injected with TR–dextran, such buds were often observed to adhere as viable bi- or trinucleate-labelled cells (see Figure 3 H). The multinucleate progeny resulting from these processes did not label with tritiated thymidine or undergo division. Figure 4 Analysis of Nuclear Migration and Fragmentation by Time-Lapse Microscopy (A) Single frames illustrating the migration of three nuclei (yellow arrows) along a myofiber, of which two are incorporated into a terminal aggregate by 11.4 h. One nucleus (green arrow) remained stationary during this period. (B) Single frames illustrating the production of viable multinucleate fragments from a myofiber. Note the presence of a trinucleate aggregate (arrowed green) that separates after lateral breakage of the fiber (0 min, arrowed yellow). This fragment subsequently extends cytoplasmic processes (14.3 and 15.4 h) and migrates over the culture substratum. Series (A) and (B) begin at 6 h after plating. Scale bars: (A) 50 μm; (B) 200 μm. Inhibition by taxol. In view of the evidence that implicates microtubules as a target for myoseverin, we stained the cultures with antibody to β-tubulin. Although tubulin was polymerised in microtubules parallel to the axis of the fibers, the regions of nuclear aggregation were associated with depolymerised tubulin ( Figure 5 A). In order to assess the functional relevance of depolymerisation, we exposed the cultures to 2 μM taxol, an agent that stabilises microtubules and inhibits division of mononucleate urodele cells without effects on cell viability or adhesion of myofibers to the culture substrate. Whereas 80% of the control fibers showed the morphologies associated with plasticity, as described above (see Materials and Methods for the criteria), only 16% were observed in the case treated with taxol ( Figure 5 B), although the total number of adherent cells was unaffected. This suggests that localised depolymerisation of microtubules may be a significant target for the regulation of these responses. Figure 5 Plasticity and Microtubule Depolymerization (A) The distribution of microtubules surrounding a multinucleate aggregate on a myofiber, as analysed by staining with anti-β-tubulin. Note the relatively disordered state of the tubulin (arrowed) in the vicinity of the nuclei. The fiber was stained at 48 h after plating. Scale bar, 50 μm. (B) Taxol inhibits the activation of myofibers after dissociation. Myofibers were dissociated and cultured in the presence of taxol as described in the Materials and Methods . The number of active fibers was determined as described. Expression of Msx1 The cultures were reacted with a digoxygenin-substituted antisense riboprobe to axolotl Msx1 . The mononucleate cells and inactive myofibers showed little or no reactivity, but active fibers showed strong reactivity with the probe in the vicinity of nuclear aggregations ( Figure 6 A and 6 B). Several control probes were negative on both classes of fibers, while the quiescent fibers as well as the active ones were reactive to antisense probes to urodele EF1a and Nrad ( Figure 6 C). In view of the relationship between Msx1 expression and plasticity in mouse myotubes ( Odelberg et al. 2000 ), the expression in the active fibers is suggestive of a role in the endogenous urodele programme. Figure 6 Analysis of mRNA Expression in Myofibers at 48 h by In Situ Hybridisation (A) Expression of Msx1 mRNA in active myofiber. Note the accumulation of reaction product around nuclear aggregates (arrowed). (B) Absence of significant Msx1 mRNA expression in a quiescent fiber. This image is taken from the same culture as (A). (C) Expression of NRad mRNA in nuclei of quiescent fibers (arrowed). Comparable intensity was observed for NRad expression by active fibers. (D) Expression of Msx1 mRNA in nuclei of fibers (arrowed) made quiescent by culture in taxol. Note the difference in Msx1 expression levels between the taxol-induced inactive fibers and normal quiescent myofibers in (B). Scale bar, 50 μm. In an initial investigation of this possibility, cultures were arrested as before by treatment with taxol, followed by reaction with the Msx1 antisense probe. In parallel control cultures, only 4.2% of the inactive fibers ( n = 404) showed any reaction with the probe, whereas 51% of all myofibers ( n = 820) showed reactivity. After treatment with taxol, 56.3% of inhibited fibers ( n = 765) were positive, whereas 63% of all myofibers ( n = 901) showed expression of Msx1 . It is clear, therefore, that the arrest of nuclear mobilisation does not prevent the early expression of Msx1 in activated myofibers. These data are consistent with an upstream role for Msx1 in the activation of plasticity, but direct evidence for functional activity has come from antisense perturbation. Activity of Morpholino-Substituted Oligonucleotides In order to evaluate the uptake of morpholino-substituted oligonucleotides, larval myofibers were dissociated as usual in the presence of 10 μM biotinylated morpholinos or underivatised morpholinos. The cells were cultured for 48 h and then analysed by a detection procedure involving tyramide signal amplification (see Materials and Methods ). Approximately 70%–90% of the fibers showed uptake of the biotinylated oligonucleotides in different experiments ( Figure 7 A), and no signal was detectable in the absence of oligonucleotide or with underivatised morpholinos ( Figure 7 B). The cells are thus effectively loaded by dissociation in the presence of morpholinos. Figure 7 Analysis of the Functional Role of Msx1 Expression by Exposure to Morpholino Antisense Oligonucleotides (A and B) Uptake of morpholino by myofibers. Myofibers were dissociated in the presence of biotinylated (A) or control (B) morpholinos and analysed by tyramide signal amplification at 24 h after plating. Note the positive signal in (A), dependent on the presence of biotin moiety. In three different experiments 70%–90% of the fibers were loaded as determined with this assay. Scale bar, 50 μm. (C) Functional effect of loading various morpholinos. Note that loading Msx1 antisense leads to a specific decrease in the proportion of active fibers relative to controls. (D–G) Staining of myofibers with antibody to Msx1 protein. (D and E) Fluorescence micrograph of a nucleus in a quiescent myofiber stained with Hoechst for DNA (D) and Msx1 protein (E). (F and G) Fluorescence micrograph of a nucleus in an active myofiber stained for DNA (F) and Msx1 protein (G). These images (D–G) were taken from the same culture. Scale bar, 20 μm. (H) Distribution of fluorescence intensity of nuclei in myofibers after staining with antibody to Msx1. The distributions for control active fibers and control quiescent fibers were determined for cells in the same culture and are significantly different (ANOVA, p < 0.001 at 95% confidence level). The distribution for antisense-treated quiescent fibers is not significantly different from that for control quiescent fibers. Limb tissue was dissociated in the presence of control morpholinos or a morpholino antisense reagent directed at the translation initiation sequence of axolotl Msx1 . The resulting cultures were analysed in parallel for the proportion of active fibers and the antisense reagent reproducibly and specifically decreased this by 60%–70% ( Figure 7 C). This led, as expected from the mechanism of such reagents, to the presence of inhibited fibers that expressed Msx1 mRNA after in situ hybridisation, and the proportion of such Msx1 positive and inhibited cells was increased by 5-fold relative to parallel cultures incubated with control oligonucleotides. The myofiber cultures were stained by indirect immunofluorescence with a rabbit antibody to Msx1 in order to evaluate the level of expression of the homeoprotein in the nuclei. There was a significant difference is staining of nuclei between active and quiescent fibers in the same culture ( Figure 7 D– 7 G). The level of expression in nuclei of different fibers was estimated by quantitative image analysis, and the distribution of intensities is shown in Figure 7 H. There was a significant difference in the fluorescence intensity of nuclei in active and quiescent fibers in control cultures ( Figure 7 H), consistent with the difference in mRNA levels observed by in situ hybridisation (see Figure 6 A and 6 B). The distribution of intensities for nuclei in quiescent fibers in parallel cultures treated with antisense oligonucleotides to Msx1 was not significantly different from the control quiescent distribution ( Figure 7 H). It should be noted that more than half of the quiescent fibers were inhibited as a result of the antisense treatment, thus indicating that the antisense distribution reflects a significant decrease in protein expression in the nucleus relative to active fibers. We conclude that expression of a critical level of Msx1 protein is necessary for the fibers to exhibit plasticity in this culture system. Discussion The plasticity of isolated urodele myofibers as described here has not been observed in previous work on dissociated mouse myofibers ( Rosenblatt et al. 1995 ; Blaveri et al. 1999 ). These apparently retain their morphological identity in culture without undergoing viable fragmentation or cellularisation. In preliminary work on myofibers dissociated from the forelimb of Xenopus tadpoles (stages 56–63), we have observed fragmentation comparable to that described here for fibers of the larval salamander, but no cellularisation. It is possible, therefore, that there is a gradation in the degree of plasticity after dissociation, and this may be related to the ability to undergo reversal during regeneration, although more work is required to investigate these comparative issues. It is interesting that the mononucleate progeny of cellularisation were observed to reenter the cell cycle, while multinucleate fragments retained the postmitotic arrest of the parental fibers. At least half of the salamander fibers were activated after dissociation and could be scored by morphological criteria as an index of plasticity, as well as by analysis of gene expression in situ. The occurrence of cellularisation did not reflect the activation of adherent mononucleate cells since the injection of a nontransferable tracer into the fibers resulted in labelling of the mononucleate progeny, and furthermore the mobilisation of nuclear aggregates occurred without any detectable S-phase reentry. It is probable that the process of enzymatic and mechanical dissociation mimics the activation events after amputation, either in terms of mechanical factors sensed by the fibers or the release of signals from the tissue or matrix. Earlier experiments on microinjected fibers in the larval tail have explored the stimuli required to trigger cellularisation and concluded that activation apparently required both ‘clipping’ at the end of the fiber as well as tissue injury in the vicinity ( Echeverri et al. 2001 ). It has also been reported that crude extracts from early regenerates of the adult newt limb are able to induce cellularisation of newt and mouse myotubes in culture ( McGann et al. 2001 ). The precise nature of the signal(s) that couples tissue injury to activation of this response remains an important subject for future investigation, particularly as it may be a key difference between urodeles and mammals. One striking consequence of fiber activation is the appearance of the Msx1 transcript, and our work strongly supports the hypothesis that Msx1 is a pivotal regulator of plasticity in differentiated cells. Although taxol treatment is able to block the internal reorganisation in activated fibers, it does not inhibit the induction of Msx1 , suggesting that microtubule depolymerisation, while being a direct target of myoseverin ( Rosania et al. 2000 ), may also be a downstream target for regulation by Msx1 . The striated myofibers are more highly differentiated than the newt A1 myotubes employed for implantation or the C2C12 mouse myotubes used to assay myoseverin and Msx1 . The events of cellularisation, cleavage, or budding off from myofibers are preceded by migration of nuclei to generate local concentrations, reminiscent of the events leading to formation of the neuromuscular junction ( Merlie and Sanes 1985 ; Englander and Rubin 1987 ), although mouse myotubes seem to undergo lateral breakage without such reorganisation ( Rosania et al. 2000 ). This migration is inhibited by taxol, and nuclear migration in other contexts is dependent on microtubule function ( Morris 2003 ). All of the events described for the myofibers occur without entry into S-phase, as determined previously for cellularisation of myotubes after implantation ( Velloso et al. 2001 ). The formation of mononucleate cells is followed by rapid division and loss of myosin expression, and these cells are presumably the culture equivalent of muscle-derived blastemal cells. The activity of the Msx1 gene has recently been implicated in digit tip regeneration in fetal and neonatal mice by comparing regeneration in normal and Msx1 mutant animals ( Reginelli et al. 1995 ; Han et al. 2003 ). It has also been shown that transgenic expression of an activated Msx1 protein can induce tail regeneration in larval Xenopus during the refractory period between stages 45 and 47 ( Beck et al. 2003 ). This evidence, taken in conjunction with the present study and that of Odelberg et al. (2000) , indicates that this gene is an important regulator of regeneration. Various activities have been associated with the protein, including a role as a repressor of transcription (reviewed in Bendall and Abate-Shen 2000 ), for example, of various myogenic differentiation genes in C2C12 myotubes ( Odelberg et al. 2000 ) and also as a positive regulator of genes that promote cell cycling such as cyclin D ( Hu et al. 2001 ). Our analysis of the myofiber cultures provides evidence for its ability to mobilise a postmitotic cell, for example, by nuclear migration and cellularisation, without S-phase reentry in the syncytium, and this suggests a different aspect of its activity as a regulator. Studies on mammalian myotubes should continue to be informative, while the present system, with its ready incorporation of antisense oligonucleotides, should be helpful for relating such studies to the endogenous programme of urodele regeneration. This in turn should assist the long-term goal of promoting the reversal of cellular differentiation as a strategy for mammalian regeneration ( Chargé and Rudnicki 2004 ). Materials and Methods Tissue dissociation and culture of myofibers The forelimbs and hind limbs of the larval spotted salamander ( A. maculatum ) or axolotl ( A. mexicanum ) (3–5 cm size) were removed, the epidermis was peeled off, and the tissue was rinsed in serum-free amphibian MEM (AMEM) ( Ferretti and Brockes 1988 ) prior to dissociation for 3 h at 26 °C in PBS containing 0.15% collagenase (Type 1A, Sigma, St. Louis, Missouri, United States), 0.8% Dispase II (Roche, Basel, Switzerland), 0.15% crystalline bovine serum albumin, 0.3% D-glucose, and 0.15 mg/ml DNase I (Roche). After 30 min of incubation, the tissues were gently triturated through a fire-polished glass pasteur pipette to aid the detachment of myofibers from the bone. After incubation, the suspension was triturated several times, centrifuged at 400 rpm for 10 min, resuspended in AMEM, and filtered through a 35 μm sieve (VWR International, Poole, United Kingdom) to remove most of the mononucleate cells. The retentate was rinsed with AMEM and plated on 35 mm Falcon Primaria (Becton-Dickinson, Palo Alto, California, United States) tissue culture dishes. Cultures were maintained at 25 °C with 2.5% CO 2 in a humidified incubator as described elsewhere ( Ferretti and Brockes 1988 ). After attachment of the myofibers to the culture dish by overnight incubation, the culture media was replaced with serum-free AMEM or AMEM supplemented with 10% foetal bovine serum. Labelling and implantation of myofibers Myofibers were dissociated as above, retained in suspension in a sterile bacteriological dish (Bibby Sterilin, Stone, United Kingdom), and incubated with 10 μM CellTracker Orange CMTMR (Molecular Probes, Eugene, Oregon, United States) for 30 min at 25 °C. The labelling was terminated by addition of 10% AMEM, and the cells were incubated for 45 min at 25 °C to permit enzymatic activation of the dye. The cell suspension was diluted several fold to allow observation of myofibers at low density. The forelimbs of axolotl larvae (7–10 cm size) were amputated at mid humerus level under tricaine (0.1%) anaesthesia 48 h before injection of labelled myofibers (see Figure 2 A). The animals were anaesthetized, and the forelimbs were positioned under a stereo zoom microscope. The myofiber suspension was placed under inverted microscope, and the myofibers were drawn into a glass micropipette (30 μm tip diameter) using an oil-driven manual microinjector (Sutter Instruments, Novato, California, United States) mounted on a Narishige (Tokyo, Japan) MMO-1 micromanipulator. The skin was punctured with a tungsten needle in order to introduce the blunt end of the micropipette. Three to eight myofibers were picked, examined carefully to verify the absence of any adherent mononucleate cells, and injected into each limb regenerate. Contralateral limbs were mock injected with medium from the suspension. The regenerates were removed at 48 h and 96 h after injection, fixed in 4% paraformaldehyde (PFA), and processed ( Kumar et al. 2000 ). Serial longitudinal sections of 60 μm were cut on a cryostat (Leica, Solms, Germany), air dried, dehydrated in PBS, and counterstained with 2.5 μM Sytox Green (Molecular Probes). The sections were observed under epifluorescence on an Axiophot microscope (Zeiss, Jena, Germany). Microinjection of cultured myofibers with conjugated dextran Myofibers were incubated in AMEM containing 2,3-butanedionne monoxime (BDM) (4 mM) for 30 min to prevent contraction of the myofibers ( Bettencourt-Dias et al. 2003 ) and maintained in the same medium during microinjection. The culture dishes were placed under a Zeiss Axiovert microscope and microinjected with TR-conjugated dextran (TR–dextran, 70 kDa; Molecular Probes). The medium was changed immediately after injection and the cultures were returned to the incubator. To identify the labelled myofibers and their mononucleate progeny, cultures were counterstained in Syto13 (200 nM; Molecular Probes) live nucleic acid stain for 30 min and observed under fluorescence microscope with a dual band pass (FITC/TRITC) filter. Live imaging of myofiber plasticity To record the coordinates of the myofibers in culture, the dish was scored underneath with a scalpel, and cells in each grid square were observed daily and images were acquired with a CCD camera (Sony, Tokyo, Japan). For time-lapse microscopy, myofiber cultures were placed under an Axiovert microscope fitted with an incubation chamber maintained at 26 °C and 3% CO 2 , and phase contrast or variable relief contrast (VAREL) (Zeiss) images were acquired using a digital camera controlled through Image-Pro Plus software (Media Cybernetics, Silver Spring, Maryland, United States). A sequence gallery was created using Image Pro-Plus and images of interest were selected, digitally enhanced, and processed in Adobe Photoshop 6.0 (Adobe, San Jose, California, United States). [ 3 H]thymidine labelling. Myofibers were incubated in 1 μCi/ml [ 3 H] thymidine (Amersham Biosciences, Little Chalfont, United Kingdom) for 24 h, fixed in 1% glutaraldehyde, and processed for autoradiography ( Velloso et al. 2000 ). DNA cytometry DNA content in myofiber nuclei and TR–dextran-labelled mononucleate progeny was measured quantitatively after fixation and staining of the nuclei with Hoechst 33258 (2 μg/ml; Sigma). Baseline values for nuclear DNA content in cycling axolotl mononucleate cells were measured in parallel after incorporation of 5-deoxy-2′-bromouridine (BrdU) (1 μM; see Tanaka et al. 1997 ). The BrdU-labelled cells were processed for double immunofluorescence with monoclonal antibody against BrdU and rabbit antibody to anti-phosphohistone H3 ( Velloso et al. 2000 ; Bettencourt-Dias et al. 2003 ). The nuclei were counterstained with Hoechst dye as above. All images were acquired using 12-bit cooled CCD camera (Photonic Sciences, Robertsbridge, United Kingdom), maintaining camera and microscope settings identical between various samples, corrected for uneven illumination and background using software functions, and processed using classification and measuring routines in Image-Pro Plus software. Scoring of plasticity in myofibers A viable nucleic acid stain such as Syto 13 or Hoechst 33342 (Molecular Probes) was routinely used in cultures to visualize and score the myofibers. The quiescent or inactive myofiber nuclei were aligned along the fiber (see Figures 1 and 7 D), and the cell did not show any cytoplasmic extensions from the axis. The nuclei in active fibers moved along the axis of the fiber to form aggregations that were localized either in the middle or towards the end of the cell (see Figure 3 A, 3 B, and 3 G). In most cases, this resulted in formation of localised cytoplasmic protrusions in the vicinity of the nuclei. Myofibers were classified and counted based on the above criteria. Taxol inhibition assay. Dissociated myofibers were plated in medium containing taxol (2 μM; Sigma). Parallel control cultures were incubated in vehicle (DMSO) in a similar way. The cultures were fixed at 48 h after treatment and processed for tubulin immunofluorescence or in situ hybridisation. Functional assay for Msx-1 using morpholino antisense oligonucleotides Morpholino-based antisense oligonucleotides of 25 oligomere were prepared to target the translation initiation site of axolotl Msx1 gene (5′-CGGTCTGCATCCTCTGCTTGCTTAG-3′) by Gene Tools Inc. (Corvallis, Oregon, United States). Invert control oligos of Msx1 (5′-GATTGCTTCGTCTCCTAGCTCTGGC-3′) and standard control oligos (5′-CCTCTTACCTCAGTTACAATTTATA-3′) from the supplier were used as controls. 3′-Biotin-end-labelled standard control oligos were used for evaluating the uptake of morpholino oligonucleotides by myofibers. Oligo stock solutions were prepared according to guidelines from the manufacturer and stored at 4 °C. The morpholino oligos were added at a concentration of 10 μM to the dissociation cocktail and the myofibers were dissociated as described. After purification of myofibers, fresh morpholino oligos were added to the culture medium. After sequential washes to remove any adherent morpholino ( McKeon et al. 2001 ), cultures were fixed at 48 h after treatment, prior to analysis. Detection of morpholino uptake by immunofluorescence. Tyramide signal amplification (PerkinElmer Life Sciences Inc., Wellesley, Massachusetts, United States) coupled with enzyme-linked immunofluorescence (ELF97, Molecular Probes) was employed to localize the uptake of morpholino oligos in cultured myofibers. Myofiber cultures were fixed at 48 h in 0.5% PFA containing 0.05% glutaraldehyde for 15 min on ice. The fixative was replaced with freshly made 0.1% NaBH 4 solution and incubated for 5 min. The manufacturer's protocol was employed for TSA amplification with the ELF97 modification. The samples were developed in ELF reaction buffer under fluorescence microscope for 10–20 s and images were acquired using a cooled digital camera. In situ hybridisation The axolotl Msx1 cDNA (a kind gift from H. Ide, Tohoku University, Sendai, Japan) was cloned into Bluescribe vector (Stratagene, La Jolla, California, United States), and probes were generated as described elsewhere ( Koshiba et al. 1998 ). A 0.7 kb axolotl EF-1α fragment (kindly provided by D. Gardiner and S. Bryant, University of California, Irvine, United States) was cloned into PCR II vector (Invitrogen, Carlsbad, California, United States) and linearised with XhoI (antisense), and a riboprobe was generated with SP6 RNA polymerase. Newt Rad ( NRad , a gift from K. Yoshizato, Hiroshima University, Hiroshima, Japan) probe was generated from a fragment of approximately 400 bp from Bluescribe vector after linearising with either HindIII (antisense) or EcoRI (sense), and riboprobes were synthesized using T3 and T7 RNA polymerase respectively ( Shimizu-Nishikawa et al. 2001 ). Axolotl EF1α and NRad probes were used as positive controls, while neomycin ( Cash et al. 1998 ), NRad sense, and Msx1 sense probes served as negative controls. For in situ hybridisation, the myofiber cultures were incubated in BDM (4 mM), fixed in chilled 1% glutaraldehyde for 15 min, postfixed in 4% PFA, and washed in 0.3% PBT. In situ hybridisation was essentially as described elsewhere ( Kumar et al. 2000 ), with minor modifications in the hybridisation temperature. Antibodies and immunofluorescence Myofiber cultures were routinely fixed in ice-cold 0.5% PFA containing 0.05% glutaraldehyde for 10 min on ice. For β-tubulin staining, 5 μM Taxol (Sigma) was incorporated into the fixative. After fixation, the culture was treated with freshly prepared 0.1% NaBH 4 for 5 min to reduce nonspecific fluorescence. The samples were post-fixed in ice-cold methanol at −20 °C for 10 min, washed three to four times in 0.3% PBT, and blocked in PBT containing 10% goat serum. The primary antibodies were to MHC and titin, and BrdU monoclonal antibody and rabbit polyclonal antibodies to phosphohistone H3, were all as described elsewhere ( Tanaka et al. 1997 ; Kumar et al. 2000 ; Velloso et al. 2000 ; Bettencourt-Dias et al. 2003 ). For localisation of β-tubulin, the culture was fixed and washed overnight in 0.3% PBT and incubated with mouse monoclonal β-tubulin antibody (1:100; clone TUB 2.1; Sigma) overnight at 4 °C. The samples were washed extensively in GS/PBT and incubated in TR-conjugated goat anti-mouse antibody (1 μg/ml; Molecular Probes). The nuclei were counterstained with Hoechst 33258 (2 μg/ml). A rabbit polyclonal antibody raised against the full-length mouse Msx1 homeoprotein was used to detect expression of Msx1 protein (BAbCO, Richmond, California, United States). When a full-length expression construct of axolotl Msx1 was expressed in mouse cells by transient transfection, the antibody gave strong and specific staining of nuclei in transfected cells ( Figure S2 ). The samples were fixed and processed as before and incubated with Msx1 antibody (1:1000) overnight at 4 °C. After several washes, the cultures were incubated with FITC-conjugated goat anti-rabbit antibodies (1:100; DakoCytomation, Cambridgeshire, United Kingdom), and the nuclei were counterstained with Hoechst. A control rabbit polyclonal antibody was processed in parallel to obtain a baseline value for quantitative fluorescence measurements on immunostained nuclei. The myofiber cultures stained for β-tubulin, MHC, or titin, or cultures injected with TR–dextran were observed under confocal laser scanning microscope (Leica). The images were acquired as z -stacks, and composite maximum projection images were generated through Leica software. Samples stained for Msx1 protein were observed under a Zeiss Axioplan microscope and images were acquired with an Axiocam digital camera. The fluorescence intensity in myofiber nuclei was measured in Axiovision software (Zeiss), and the data were analysed by one-way analysis of variance (ANOVA) followed by multiple range test using Instat (Graphpad Software Inc., San Diego, California, United States). Supporting Information Figure S1 Live Striated Myofiber Dissociated from the Limb of a Larval Salamander The myonuclei incorporate Syto13 live nuclear stain. The myofiber was observed with VAREL optics at 24 h after plating. Scale bar, 100 μm. (4.1 MB TIF). Click here for additional data file. Figure S2 Expression of Newt Msx1 in Mouse PS Cells by Transient Transfection Nuclear localisation of Msx1 protein (green) was detected with a rabbit polyclonal antibody generated against the full-length mouse Msx1 homeoprotein. (5.6 MB TIF). Click here for additional data file. Video S1 Time-Lapse Video Analysis of Nuclear Migration in a Myofiber Time-lapse sequence was begun 6 h after plating of the myofiber on to a culture dish. The images were taken at 6 min intervals under 32× VAREL objective magnification. (110 KB AVI). Click here for additional data file.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC509293.xml
533858	Models of chronic obstructive pulmonary disease	Chronic obstructive pulmonary disease (COPD) is a major global health problem and is predicted to become the third most common cause of death by 2020. Apart from the important preventive steps of smoking cessation, there are no other specific treatments for COPD that are as effective in reversing the condition, and therefore there is a need to understand the pathophysiological mechanisms that could lead to new therapeutic strategies. The development of experimental models will help to dissect these mechanisms at the cellular and molecular level. COPD is a disease characterized by progressive airflow obstruction of the peripheral airways, associated with lung inflammation, emphysema and mucus hypersecretion. Different approaches to mimic COPD have been developed but are limited in comparison to models of allergic asthma. COPD models usually do not mimic the major features of human COPD and are commonly based on the induction of COPD-like lesions in the lungs and airways using noxious inhalants such as tobacco smoke, nitrogen dioxide, or sulfur dioxide. Depending on the duration and intensity of exposure, these noxious stimuli induce signs of chronic inflammation and airway remodelling. Emphysema can be achieved by combining such exposure with instillation of tissue-degrading enzymes. Other approaches are based on genetically-targeted mice which develop COPD-like lesions with emphysema, and such mice provide deep insights into pathophysiological mechanisms. Future approaches should aim to mimic irreversible airflow obstruction, associated with cough and sputum production, with the possibility of inducing exacerbations.	Introduction The global burden of disease studies point to an alarming increase in the prevalence of chronic obstructive pulmonary disease (COPD) [ 1 ] which is predicted to be one of the major global causes of disability and death in the next decade [ 2 ]. COPD is characterized by a range of pathologies from chronic inflammation to tissue proteolysis and there are no drugs specifically developed for COPD so far. Cessation of cigarette smoking is accompanied by a reduction in decline in lung function [ 3 ] and is a most important aspect of COPD management. The mainstay medication consists of beta-adrenergic and anticholinergic bronchodilators; addition of topical corticosteroid therapy in patients with more severe COPD provides may enhance bronchodilator responses and reduce exacerbations [ 4 ]. In contrast to the large amount of experimental studies on allergic asthma and the detailed knowledge that exists on mediators of allergic airway inflammation [ 5 , 6 ], much less has been conducted for COPD. More effort and resources have been directed into asthma research in comparison to COPD. The available insights into the pathogenesis and pathophysiology of asthma may help to improve research in COPD [ 7 ]. Many research centres that previously focused on asthma now also investigate mechanisms of COPD. Using molecular and genetic approaches, an increasing range of molecules has been identified that could underlie the pathogenic inflammation of chronic allergic airway inflammation [ 8 ]. Based on these findings and on new ways of administering drugs to the lungs [ 9 ], a new image of overwhelming complexity of the underlying pathophysiology of COPD has emerged (Figure 1 ). The current challenge in COPD research is to identify the role of the various mediators and molecular mechanisms that may be involved in its pathophysiology, and obtain new treatments. In addition, it is incumbent to understand the effect of smoking cessation on the pathogenetic process. Figure 1 Potential pathogenetic mechanisms involved in COPD Exogenous inhaled noxious stimuli such as tobacco smoke, noxious gases or indoor air pollution and genetic factors are proposed to be the major factors related to the pathogenesis of COPD. These factors may influence protease activity and may also lead to an imbalance between pro-inflammatory and anti-inflammatory mediators. Studying the molecular pathways in human subjects is restricted to the use of morphological and molecular assessment of lung tissues obtained at surgery or performing limited in vitro studies at one single point in time [ 10 ]. There is a need for in vivo animal models to examine more closely pathogenesis, functional changes and the effects of new compounds or treatments. However, animal models have limitations since there is no spontaneous model, and models do not necessarily mimic the entire COPD phenotype. The best model remains chronic exposure to cigarette smoke, since this is the environmental toxic substance(s) that cause COPD in man. However, other substances are also implicated such as environmental pollution due to car exhaust fumes. The present review draws attention to specific aspects of functional and structural features of COPD that need to be realized when interpreting molecular mechanisms identified in animal models of COPD. It identifies important issues related to the ongoing experimental COPD research which may in the future provide optimized COPD diagnosis and treatment. COPD Clinical features Before characterizing and discussing the different animal models of COPD which have been established so far, it is crucial to reflect that within COPD, different disease stages exist and that only some of them may be mimicked in animal models. The diagnosis of COPD largely relies on a history of exposure to noxious stimuli (mainly tobacco smoke) and abnormal lung function tests. Since COPD has a variable pathology and the molecular mechanisms are only understood to a minor extent, a simple disease definition has been difficult to establish. However, the diagnosis of COPD relies on the presence of persistent airflow obstruction in a cigarette smoker [ 4 ]. A classification of disease severity into four stages has been proposed by the GOLD guidelines based primarily on FEV1 [ 4 ]. The staging on the basis of FEV1 alone as an index of severity for COPD has been criticised. A composite measure essentially based on clinical parameters (BODE) has been shown to be better at predicting mortality than FEV1 [ 11 ]. The natural history of COPD in terms of evolution of FEV1 remains unclear and the temptation is to regard the stages as evolving from Stage 0 to Stage 4. Just as many smokers do not develop COPD, it is possible that the disease may not progress from one stage to the next. Some patients with severe COPD are relatively young and it is not clear if early stages of their disease are similar to those found in patients with mild COPD. COPD is a heterogeneous disease and different possible outcomes may occur at each of the stages. Experimental modeling of each stage of severity may be a way of providing an answer to this issue. Animal models may also help to provide a better classification of severity by correlating biochemical, molecular and structural changes with lung function and exercise tolerance. Pathophysiology The presence of airflow obstruction which has a small reversible component, but which is largely irreversible is a major feature of COPD as indicated by the Global Initiative for Chronic Obstructive Lung Disease (GOLD) guidelines [ 4 ]. It is proposed to be the result of a combination of small airways narrowing, airway wall inflammation [ 12 ] and emphysema-related loss of lung elastic recoil [ 13 , 14 ]. These features differ to a large extent to findings observed in bronchial asthma (Table 1 ) where airflow obstruction is usually central, while involvement of the small airways occurs in more severe disease. The degree of airflow obstruction in COPD can be variable, but loss of lung function over time is a characteristic feature. Ideally, the development of airflow obstruction which is largely irreversible but has a small reversible component should be a feature of animal models of COPD, but this has not been reproduced so far. One of the important limitations of animal models of COPD is the difficulty in: reproducing small airways pathology particularly when working in small animals, particularly the mouse and rat where there are few levels of airway branching. This is a problem inherent to small laboratory animal models but provides an advantage for developing models in larger animals such as the pig or sheep. Part of the problem of analyzing small airways is also due to the lack of sophistication of lung function measurements, particularly in mice, but there has been recent development in the methodology of lung function measurement [ 15 ]. A new ex-vivo method of analyzing the airway periphery is by the technique of precision cut lung slices combined to videomorphometry [ 16 , 17 ]. Table 1 Currently known phenotype differences between COPD and asthma Feature COPD Asthma Limitation of Airflow Largely irreversible Largely reversible Parenchymal integrity destruction intact Bronchial Hyperresponsiveness Variable (small) significant Steroid response reduced or absent present In addition to pulmonary alterations, other organ systems may be affected in COPD [ 18 ]. Systemic effects of COPD include weight loss, nutritional abnormalities and musculoskeletal dysfunction. These systemic manifestations will gain further socioeconomic importance with an increasing prevalence of COPD in the next years [ 19 ]. Therefore, these systemic effects should be present in animal models of COPD and further analysis of mechanisms underlying these systemic effects in experimental models may help to optimize disease management. Inflammatory cells An important feature of COPD is the ongoing chronic inflammatory process in the airways as indicated by the current GOLD definition of COPD [ 4 ]. There are differences between COPD and asthma: while mast cells and eosinophils are the prominent cell types in allergic asthma, the major inflammatory cell types in COPD are different (Table 2 ) [ 20 - 22 ]. Table 2 Differences in inflammatory cells between COPD and asthma. Ranked in relative order of importance. COPD Asthma Neutrophils Eosinophils Macrophages Mast cells CD8-T-lymphocytes CD4-T-lymphocytes Eosinophils (exacerbations) Macrophages, Neutrophils Neutrophils play a prominent role in the pathophysiology of COPD as they release a multitude of mediators and tissue-degrading enzymes such as elastases which can orchestrate tissue destruction and chronic inflammation [ 8 , 23 ]. Neutrophils and macrophages are increased in bronchoalveolar lavage fluid from cigarette smokers [ 24 ]. Patients with a high degree of airflow limitation have a greater induced sputum neutrophilia than subjects without airflow limitation. Increased sputum neutrophilia is also related to an accelerated decrease in FEV 1 and sputum neutrophilia is more prevalent in subjects with chronic cough and sputum production [ 25 ]. The second major cell type involved in cellular mechanisms are macrophages [ 26 ]. They can release numerous tissue-degrading enzymes such as matrix metalloproteinases (MMPs). In an animal model of tobacco smoke-induced tissue matrix degradation, not only neutrophil enzymes but also macrophage-derived enzymes such as MMP-12 are important for the development of emphysema-like lesions [ 27 ]. A further key enzyme is the macrophage metalloelastase which was reported to mediate acute cigarette smoke-induced inflammation via tumor necrosis factor (TNF)-alpha-release [ 28 ]. Neutrophils and macrophages can communicate with other cells such as airway smooth muscle cells, endothelial cells or sensory neurons, and release inflammatory mediators that induce bronchoconstriction [ 29 ], airway remodelling [ 30 ], and mucin gene induction and mucus hypersecretion involving the induction of mucin genes [ 31 - 33 ]. Lymphocytes are also involved in cellular mechanisms underlying COPD [ 34 , 35 ]. Increased numbers of CD8-positive T-lymphocytes are found in the airways of COPD patients [ 21 , 22 ] and the degree of airflow obstruction is correlated with their numbers [ 36 ]. However, the T-cell associated inflammatory processes largely differ from those in allergic asthma, which is characterized by increased numbers of CD4-positive T-lymphocytes [ 7 , 37 ] (Table 2 ). Although eosinophils may only play a major role in acute exacerbations of COPD [ 38 ], their presence in stable disease is an indicator of steroid responsiveness [ 39 - 41 ]. Different inflammatory cell types have also been characterized in airway tissues. Epithelial neutrophilia has been seen in proximal and distal airways of patients with COPD [ 42 , 43 ]. The airway wall beneath the epithelium shows a mononuclear inflammation with increased macrophages and T cells bearing activation markers [ 20 , 36 ] Di Stefano 1996;. An excess od CD8+ T cells are particularly observed in central airways, peripheral airways and parenchyma [ 20 , 43 ]. In the small airways from patients with stage 0 to (at risk) stage 4 (very severe) COPD, the progression of the disease is strongly associated with the accumulation of inflammatory exudates in the small airway lumen and with an increase in the volume of tissue in the airway wall [ 10 ]. Also, the percentage of airways containing macrophages, neutrophils, CD4 cells, CD8 cells, B cells, and lymphoid follicle aggregates and the absolute volume of CD8+ T-cells and B cells increased with the progression of COPD [ 10 ]. The changes are also most likely associated with an induction of mucin gene expression [ 44 ]. The presence of increased numbers of B cells begs the question regarding the role of these cells in the pathophysiology of COPD. In the airway smooth muscle bundles in smokers with COPD, increased localisation of T- cells and neutrophils has been reported, indicating a possible role for these cells interacting with airway smooth muscle in the pathogenesis of airflow limitation [ 45 ]. Mechanisms of COPD On the basis of the different pathophysiological mechanisms illustrated in Fig. 1 , different animal models have been developed in past years. Protease-antiprotease imbalance An imbalance between protease and antiprotease enzymes has been hypothesized with respect to the pathogenesis of emphysema [ 46 ]. This concept derives from early clinical observations that alpha1-antitrypsin-deficient subjects develop severe emphysema and the role of protease-antiprotease imbalance was later demonstrated in animal models of COPD [ 47 , 48 ]. Although alpha1-antitrypsin-deficiency is a very rare cause of emphysema [ 49 , 50 ], it points to a role of proteases and proteolysis [ 51 , 52 ]. Neutrophil elastase-deficient mice were significantly protected from emphysema-development induced by chronic cigarette smoke [ 48 ]. Depletion of the macrophage elastase gene also led to a complete protection from emphysema induced by cigarette smoke [ 47 ]. Each of these elastases inactivated the endogenous inhibitor of the other, with macrophage elastase degrading alpha1-antitrypsin and neutrophil elastase degrading tissue inhibitor of metalloproteinase-1 [ 48 ]. In tobacco smoke exposure-induced recruitment of neutrophils and monocytes was impaired in elastase gene-depleted animals and there was less macrophage elastase activity due to a decreased macrophage influx in these animals. Thus, a major role for neutrophil elastase and macrophage elastase in the mediation of alveolar destruction in response to cigarette smoke has been shown [ 47 , 48 ]. This experimental evidence derived from animal models points to an important pathogenetic role for proteases that correlates well with the imbalance of proteases present in human COPD. However, many pathways of tissue destruction can be found in animal models that lead to a picture similar to human disease, and it is important to examine whether these mechanisms are operative in the human disease itself. Oxidative stress Oxidative stress arising from inhaled noxious stimuli such as tobacco smoke or nitrogen dioxide may be important cause of the inflammation and tissue damage in COPD. This potential mechanism is supported by clinical reports of increased levels of oxidative stress indicators in exhaled breath condensates of COPD patients [ 53 - 55 ]. Apart from elevated levels of 8-isoprostane [ 55 ], nitrosothiol levels were increased in COPD patients [ 56 - 58 ]. Studies in a mouse model of tobacco smoke-induced COPD also demonstrated the presence of tissue damage due to oxidative stress [ 59 ]. These changes could be blocked by superoxide dismutase [ 60 ]. Oxidative stress has also been implicated in the development of corticosteroid resistance in COPD. Mediators Many mediators have been identified which may contribute to COPD pathogenesis [ 8 ]. As in bronchial asthma, pro- and anti-inflammatory mediators of inflammation such as tachykinins [ 61 ], vasoactive intestinal polypeptide (VIP) [ 62 ], histamine [ 63 ], nitric oxide [ 64 , 65 ], leukotrienes [ 66 ], opioids [ 67 ] or intracellular mediators such as SMADs [ 68 , 69 ] have been implicated. The balance of histone acetylases and deacetylases [ 70 ] is a key regulator of gene transcription and expression by controlling the access of the transcriptional machinery to bind to regulatory sites on DNA. Acetylation of core histones lead to modification of chromatin structure that affect transcription, and the acetylartion status depends on a balance of histone deacetylase and histone acetyltransferase. This is also likely to play a role in the regulation of cytokine production in COPD. Cigarette smoke exposure led to altered chromatin remodelling with reduced histone deacetylase activity with a resultant increase in transcription of pro-inflammatory genes in lungs of rats exposed to smoke, linked to an increase in phosphorylated p38 MAPK in the lung concomitant with an increased histone 3 phospho-acetylation, histone 4 acetylation and elevated DNA binding of NF-kappaB, and activator protein 1 (AP-1) [ 70 ]. In addition, oxidative stress has also been shown to enhance acetylation of histone proteins and decrease histone deacetylase activity leading to modulation of NF-κB activation [ 71 ], similar to the effect of cigarette smoke. A Th2 cytokine that has been proposed to be implicated in the pathophysiology of COPD is IL-13. It is also overexpressed and related to the pathogenesis of the asthmatic Th2 inflammation and airway remodelling process [ 72 ]. The effects of IL-13 in asthma have been elucidated in a series of experiments that demonstrated the an airway-specific constitutive overexpression of IL-13 leads to a process of airway remodelling with subepithelial fibrosis and mucus metaplasia combined with an eosinophil-, lymphocyte-, and macrophage-rich inflammation and increased hyperresponsiveness [ 73 ]. Since asthma and COPD pathogenesis may be linked, similar mechanisms may contribute to the development and progression of both diseases [ 74 ]. In this respect, IL-13 may also play a role in COPD since the inducible overexpression of IL-13 in adult murine lungs leads to alveolar enlargement, lung enlargement and an enhanced compliance and mucus cell metaplasia [ 75 ] with activation of MMP-2, -9, -12, -13, and -14 and cathepsins B, S, L, H, and K in this model. Parallel to protease-based and extracellular mediator-based concepts, altered intracellular pathways may also play a role in COPD. MAPK signalling pathways i.e. p38 and c-Jun N terminal kinase (JNK) [ 76 , 77 ] seem to be important signal transducers in the airways and airway-innervating neurons [ 78 - 80 ] and may therefore display an interesting target for COPD research. For some cells, the activation of p38 or JNK pathways may promote apoptosis rather than proliferation [ 81 , 82 ]. Viral infections Previous studies showed an association between latent adenoviral infection with expression of the adenoviral E1A gene and chronic obstructive pulmonary disease (COPD) [ 83 , 84 ]. It may therefore be assumed that latent adenoviral infection can be one of the factors that might amplify airway inflammation. Human data [ 35 ] demonstrating the presence of the viral E1A gene and its expression in the lungs from smokers [ 85 , 86 ], animals [ 87 ] and cell cultures [ 88 ] support this hypothesis. A small population of lung epithelial cells may carry the adenoviral E1A gene which may then amplify cigarette smoke-induced airway inflammation to generate parenchymal lesions leading to COPD. Inflammatory changes lead to collagen deposition, elastin degradation, and induction of abnormal elastin in COPD [ 89 , 90 ]. Also, latent adenovirus E1A infection of epithelial cells could contribute to airway remodelling in COPD by the viral E1A gene, inducing TGF-beta 1 and CTGF expression and shifting cells towards a more mesenchymal phenotype[ 84 ]. Genetics Since only a minority of smokers (approximately 15 to 20%) develop symptoms and COPD is known to cluster in families, a genetic predisposition has been hypothesized. Many candidate genes have been assessed, but the data are often unclear and systematic studies are currently performed to identify disease-associated genes. Next to alpha1-antitrypsin deficiency, several candidate genes have been suggested to be linked to COPD induction. Genetic polymorphisms in matrix metalloproteinase genes MMP1, MMP9 and MMP12 may be important in the development of COPD. In this respect, polymorphisms in the MMP1 and MMP12 genes, but not MMP9, have been suggested to be related to smoking-related lung injury or are in a linkage disequilibrium with other causative polymorphisms [ 91 - 93 ]. An association between an MMP9 polymorphism and the development of smoking-induced pulmonary emphysema was also reported in a population of Japanese smokers [ 94 ]. Also, polymorphisms in the genes encoding for IL-11 [ 95 ], TGF-beta1 [ 96 ], and the group-specific component of serum globulin [ 97 ] have been shown to be related to a genetic predisposition for COPD. Since it was difficult to replicate some of these findings among different populations, future studies are needed. Also, whole genome screening in patients and unaffected siblings displays a promising genetic approach to identify genes associated with COPD. Experimental models of COPD There are three major experimental approaches to mimic COPD encompassing inhalation of noxious stimuli, tracheal instillation of tissue-degrading enzymes to induce emphysema-like lesions and gene-modifying techniques leading to a COPD-like phenotype (Figure 2 ). These approaches may also be combined. Ideally a number of potential indicators for COPD which have been proposed by the GOLD guidelines should be present in animal models of COPD (Table 3 ). Since COPD definition still rests heavily on lung function measures (airflow limitation and transfer factor), it would be ideal to have lung function measurements in experimental models [ 15 ]. The challenge is in the measurement of lung function in very small mammals such as mice and since the use of the enhanced pause (Penh) in conscious mice as an indicator of airflow obstruction is not ideal [ 98 ], invasive methods remain the gold standard and these should be correlated with inflammatory markers and cellular remodelling. Figure 2 Experimental approaches to mimic COPD There are three major experimental approaches to mimic COPD or emphysema consisting of inhalation of noxious stimuli such as tobacco smoke, tracheal instillation of tissue-degrading enzymes to induce emphysema-like lesions and gene-modifying techniques leading to COPD-like murine phenotypes. Table 3 Indicators for COPD. These indicators are related to the presence of COPD and should ideally be present in animal models and available for analysis. Indicator Human features Experimental approach History of exposure to risk factors Tobacco smoke. Occupational dusts and chemicals. Indoor / outdoor air pollution Exposure-based experimental protocol Airflow obstruction Decrease in FEV 1 Lung function tests Hypersecretion Chronic sputum production Functional and morphological assessment of hypersecretion Cough Chronic intermittent or persistent cough Cough assessment Dyspnea Progressive / Persistent / worse on exercise / worse during respiratory infections Assessment of hypoxemia Emphysema Progressive impairment of lung function Morphological analysis of airspace enlargement Inhalation models – tobacco smoke A variety of animal species has been exposed to tobacco smoke. Next to guinea pigs, rabbits, and dogs, and also rats and mice have been used. Guinea pigs have been reported to be a very susceptible species. They develop COPD-like lesions and emphysema-like airspace enlargement within a few months of active tobacco smoke exposure [ 99 ]. By contrast, rat strains seem to be more resistant to the induction of emphysema-like lesions. Susceptibility in mice varies from strain to strain. The mode of exposure to tobacco smoke may be either active via nose-only exposure systems or passive via large whole-body chambers. The first species to be examined in detail for COPD-like lesions due to tobacco smoke exposure was the guinea pig [ 99 ]. Different exposure protocols were screened and exposure to the smoke of 10 cigarettes each day, 5 days per week, for a period of either 1, 3, 6, or 12 months resulted in progressive pulmonary function abnormalities and emphysema-like lesions. The cessation of smoke exposure did not reverse but stabilized emphysema-like airspace enlargement. On the cellular level, long term exposure lead to neutrophilia and accumulation of macrophages and CD4+ T-cells [ 83 , 100 ]. Latent adenoviral infection amplifies the emphysematous lung destruction and increases the inflammatory response produced by cigarette-smoke exposure. Interestingly, it was shown that the increase in CD4+ T-cells is associated with cigarette smoke and the increase in CD8+ T-cells with latent adenoviral infection [ 83 ]. Mice represent the most favoured laboratory animal species with regard to immune mechanisms since they offer the opportunity to manipulate gene expression. However, it is more difficult to assess lung function and mice tolerate at least two cigarettes daily for a year with minimal effects on body weight and carboxyhemoglobin levels. Mice differ considerably in respiratory tract functions and anatomy if compared to humans: they are obligate nose breathers, they have lower numbers of cilia, fewer Clara cells and a restriction of submucosal glands to the trachea. Next to a lower filter function for tobacco smoke, mice also do not have a cough reflex and many mediators such as histamine or tachykinins have different pharmacological effects. The development of emphysema-like lesions is strain-dependent: enlarged alveolar spaces and increased alveolar duct area are found after 3–6 months of tobacco smoke exposure in susceptible strains such as B6C3F1 mice [ 101 ]. At these later time points, tissue destruction seems to be mediated via macrophages. At the cellular level, neutrophil recruitment has been reported to occur immediately after the beginning of tobacco smoke exposure and is followed by accumulation of macrophages. The early influx of neutrophils is paralleled by a connective tissue breakdown. The early stage alterations of neutrophil influx and increase in elastin and collagen degradation can be prevented by pre-treatment with a neutrophil antibody or alpha1-antitrypsin [ 102 ]. Rats are also often used for models of COPD. However, they appear to be relatively resistant to the induction of emphysema-like lesions. Using morphometry and histopathology to assess and compare emphysema development in mice and rats, significant differences were demonstrated [ 101 ]: Animals were exposed via whole-body exposure to tobacco smoke at a concentration of 250 mg total particulate matter/m3 for 6 h/day, 5 days/week, for either 7 or 13 months. Morphometry included measurements of tissue loss (volume density of alveolar septa) and parenchymal air space enlargement (alveolar septa mean linear intercept, volume density of alveolar air space). Also, centroacinar intra-alveolar inflammatory cells were assessed to investigate differences in the type of inflammatory responses associated with tobacco smoke exposure. In B6C3F1 mice, many of the morphometric parameters used to assess emphysema-like lesions differed significantly between exposed and non-exposed animals. By contrast, in exposed Fischer-344 rats, only some parameters differed significantly from non-exposed values. The alveolar septa mean linear intercept in both exposed mice and rats was increased at 7 and 13 months, indicating an enlargement of parenchymal air spaces. In contrast, the volume density of alveolar air space was significantly increased only in exposed mice. The volume density of alveolar septa was decreased in mice at both time points indicating damage to the structural integrity of parenchyma. There was no alteration in Fischer-344 rats. Morphologic evidence of tissue destruction in the mice included irregularly-sized and -shaped alveoli and multiple foci of septal discontinuities and isolated septal fragments. The morphometric differences in mice were greater at 13 months than at 7 months, suggesting a progression of the disease. Inflammatory influx within the lungs of exposed mice contained significantly more neutrophils than in rats. These results indicated that B6C3F1 mice are more susceptible than F344-rats to the induction of COPD-like lesions in response to tobacco smoke exposure [ 101 ]. Recent work on cigarette exposure in rats indicate that this model also achieves a degree of corticosteroid resistance that has been observed in patients with COPD [ 103 , 104 ]. Thus, the inflammatory response observed after exposure of rats to cigarette smoke for 3 days is noty inhibited by pre-treatment with corticosteroids [ 70 ]. This may be due to the reduction in histone deacetylase activity, which could result from a defect in recruitment of this activity by corticosteroid receptors. Corticosteroids recruit hitone deacetylase 2 protein to the transcriptional complex to suppress proinflammatory gene transcription [ 105 ]. Modifications in histone deacetylase 2 by oxidative stress or by cigarette smoke may make corticosteroids ineffective [ 106 ]. Therefore, models of COPD that show corticosteroid resistance may be necessary and could be used to dissect out the mechanisms of this resistance. Generally, tobacco smoke exposure may be used to generate COPD features such as emphysema and airway remodelling and chronic inflammation. Although the alterations still differ from the human situation and many involved mediators may have different functional effects especially in the murine respiratory tract, these models represent useful approaches to investigate cellular and molecular mechanisms underlying the development and progression of COPD. As a considerable strain-to-strain and species-to-species variation can be found in the models used so far, the selection of a strain needs to be done with great caution. Animal models of COPD still need to be precisely evaluated as to whether they mimic features of human COPD, and their limitations must be appreciated. Findings obtained from these models may provide significant advances in terms of understanding novel mechanisms involved in COPD. Inhalation models – sulfur dioxide Sulfur dioxide (SO 2 ) is a gaseous irritant which can be used to induce COPD-like lesions in animal models. With daily exposure to high concentrations of SO 2 , chronic injury and repair of epithelial cells can be observed in species such as rat or guinea pig. The exposure to high-levels of this gas ranging from 200 to 700 ppm for 4 to 8 weeks has been demonstrated to lead to neutrophilic inflammation, morphological signs of mucus production and mucus cell metaplasia and damage of ciliated epithelial cells in rats [ 107 , 108 ]. These changes are directly dependent on the exposure to the gas: signs of mucus production and neutrophilic inflammation are almost entirely reversed within a week after termination of exposure [ 108 ]. Acute exposure to SO 2 also leads to loss of cilia and exfoliation of ciliated cells as demonstrated in SO 2 -exposed dogs using transmission electron microscopy [ 109 ]. After a longer period of exposure the epithelial layer regenerates and airway wall thickening and change in cilia structure can be observed [ 110 ]. Long-term exposure also increases in mucosal permeability both in vivo and in vitro [ 111 ]. Mucus hypersecretion is an important indicator for COPD and experimental models should encompass features of hypersecretion. After chronic exposure to SO 2 in rats, visible mucus layers and mucus plugs may sometimes be observed in the large airways [ 107 ] and an elevation of mucus content may be found in bronchoalveolar lavage fluids [ 112 ]. Parallel to these findings, there is an increase of PAS- and Alcian Blue-staining epithelial cells in chronically SO 2 exposed rats [ 113 ] but there is substantial variation present as with human COPD [ 114 ]. Tracheal mucus glands are also increased in size after SO 2 -exposure [ 115 ] and increased levels of mucin RNA can be found in lung extracts [ 112 ]. The mechanisms underlying mucus hypersecretion have not been elucidated so far and also, functional studies assessing basal and metacholine-induced secretion have not been conducted so far. Airway inflammation with cellular infiltration is an important feature of COPD. After exposure to SO 2 , increases in mononuclear and polymorphonuclear inflammatory cells are present in rat airways. However, the influx is confined to large but not small airways which are important in human COPD [ 107 ]. Even after one day of exposure, polymorphonuclear inflammatory cells are found and their influx can be inhibited with steroid treatment [ 116 ]. SO 2 -based models of COPD have also been shown to be associated with an increase in pulmonary resistance and airway hyperresponsiveness [ 107 ] and it was hypothesized that elevated levels of mucus may account for the increased responsiveness [ 117 ]. Since sensory nerve fibres may function as potent regulators of chronic inflammation in COPD by changes in the activation threshold and the release of pro-inflammatory mediators such as tachykinins [ 61 , 118 ] or CGRP [ 6 , 119 ], this class of nerve fibres was examined in a number of studies [ 120 , 121 ]. The results of these studies supported the hypothesis that rather than contributing to the pathophysiological manifestations of bronchitis, sensory nerve fibres limit the development of airway obstruction and airway hyperresponsiveness during induction of chronic bronchitis by SO 2 -exposure. In this respect, the enhanced contractile responses of airways from neonatally SO 2 -exposed capsaicin-treated rats may result from increased airway smooth muscle mass and contribute to the increased airway responsiveness observed in these animals [ 121 ]. To obtain coexisting expression of emphysema and inflammatory changes as seen in COPD, neutrophil elastase instillation and SO 2 -exposure were performed simultaneously [ 108 ]. The pre-treatment with elastase aimed to render the animals more susceptible to the inflammation induced by SO 2 . However, neither allergy-phenotype Brown Norway nor emphysematous Sprague–Dawley rats displayed an increased sensitivity to SO 2 -exposure. With regard to the observed histopathological changes, it can be concluded that SO 2 exposure leads to a more diffuse alveolar damage with a more extensive damage with destruction of lung tissue after longer exposure. Therefore, the outcome is more or less a picture of tissue destruction with close resemblance to end stages of emphysema but not a complete picture of COPD. Inhalation models – nitrogen dioxide Nitrogen dioxide (NO 2 ) is a another gas that may lead to COPD-like lesions depending on concentration, duration of exposure, and species genetic susceptibility [ 122 ]. Concentrations ranging from 50–150 ppm (94–282 mg/m3) can lead to death in laboratory animals due to extensive pulmonary injury including pulmonary oedema, haemorrhage, and pleural effusion. Short-term exposure to NO2 leads to a biphasic response with an initial injury phase followed by a repair phase. Both increased cellular proliferation and enzymatic activity occur during the repair phase. Exposure of rats to 15 ppm NO 2 for 7 days leads to an increased oxygen consumption in airway tissues. The increase in oxidative capacity reflects an increase in mitochondrial activity consistent with observations of increased DNA synthesis [ 123 ]. Exposure to 10 ppm NO 2 for more than 24 h causes damage to cilia and hypertrophy of the bronchiolar epithelium [ 124 ]. Also, exposure to 15–20 ppm NO 2 leads to a type II pneumocyte hyperplasia [ 125 , 126 ]. As with the exposure to other noxious stimuli, there is also a significant inter-species variability. In comparison to mice and rats, guinea pigs exhibit changes in lung morphology at much lower NO 2 concentrations. It was shown that a 2 ppm NO 2 3-day exposure causes increased thickening of the alveolar wall, damage to cilia and pulmonary oedema [ 127 ]. Other changes are an influx of inflammatory cells and increases in connective tissue formation [ 128 ]. There is also a significant mode of inheritance of susceptibility to NO 2 -induced lung injury in inbred mice. Susceptible C57BL/6J (B6) and resistant C3H/HeJ (C3) mice, as well as F1, F2, and backcross (BX) populations derived from them, were acutely exposed to 15 parts per million NO 2 for 3 h to determine differences [ 122 ]. Significant differences in numbers of lavageable macrophages, epithelial cells, and dead cells were found between inbred strains: distributions of cellular responses in F1 progeny overlapped both progenitors, and mean responses were intermediate. It was shown that in C3:BX progeny, ranges of responses to NO 2 closely resembled C3 mice. Ranges of cellular responses to NO 2 in B6:BX and intercross progeny were reported to overlap both progenitor and mean responses of both populations were intermediate to progenitors. Therefore, there were likely two major unlinked genes that account for differential susceptibility to acute NO 2 exposure [ 122 ]. Based on the genetic background of C57BL/6 mice, a model of long-term NO 2 exposure was recently established leading to signs of pulmonary inflammation and progressive development of airflow obstruction [ 129 ]. Inhalation models – oxidant stimuli and particulates The administration of oxidants such as ozone also causes significant lung injury with some features related to inflammatory changes occurring in human COPD [ 130 ] and this causes numerous effects in airway cells [ 131 - 135 ]. As a gaseous pollutant, ozone targets airway tissues and breathing slightly elevated concentrations of this gas leads to a range of respiratory symptoms including decreased lung function and increased airway hyper-reactivity. In conditions such as COPD and asthma, ozone may lead to exacerbations of symptoms. Ozone is highly reactive: the reaction with other substrates in the airway lining fluid such as proteins or lipids leads to secondary oxidation products which transmit the toxic signals to the underlying pulmonary epithelium. These signals include cytokine generation, adhesion molecule expression and tight junction modification leading to inflammatory cell influx and increase of lung permeability with oedema formation [ 130 ]. However, the nature and extent of these responses are often variable and not related within an individual. The large amount of data obtained from animal models of ozone exposure indicates that both ozone- and endotoxin-induced animal models are dependent on neutrophilic inflammation. It was shown that each toxin enhances reactions induced by the other toxin. The synergistic effects elicited by coexposure to ozone and endotoxin are also mediated, in part, by neutrophils. [ 136 , 137 ]. Further animal models focus on the exposure to ultrafine particles, silica and coal dust [ 138 , 139 ]. Ultrafine particles are a common component of air pollution, derived mainly from primary combustion sources that cause significant levels of oxidative stress in airway cells [ 140 , 141 ]. The animal models are predominantly characterized by focal emphysema and it was suggested that dust-induced emphysema and smoke-induced emphysema occur through similar mechanisms [ 142 ]. Exposure to diesel exhaust particles (DEP) may also lead to chronic airway inflammation in laboratory animals as it was shown to have affect various respiratory conditions including exacerbations of COPD, asthma, and respiratory tract infections [ 143 ]. Both the organic and the particulate components of DEP cause significant oxidant injury and especially the particulate component of DEP is reported to induce alveolar epithelial damage, alter thiol levels in alveolar macrophages (AM) and lymphocytes, and induce the generation of reactive oxygen species (ROS) and pro-inflammatory cytokines [ 144 ]. The organic component has also been shown to generate intracellular ROS, leading to a variety of cellular responses including apoptosis. Long-term exposure to various particles including DEP, carbon black (CB), and washed DEP devoid of the organic content, have been shown to produce chronic inflammatory changes and tumorigenic responses [ 144 ]. The organic component of DEP also suppresses the production of pro-inflammatory cytokines by macrophages and the development of Th1 cell-mediated mechanisms thereby enhancing allergic sensitization. The underlying mechanisms have not been fully investigated so far but may involve the induction of haeme oxygenases, which are mediators of airway inflammation [ 145 ]. Whereas the organic component that induces IL-4 and IL-10 production may skew the immunity toward Th2 response, the particulate component may stimulate both the Th1 and Th2 responses [ 146 ]. In conclusion, exposure to particulate and organic components of DEP may be a helpful approach to simulate certain conditions such as exacerbations. Also, the development of lung tumours after long term exposure may be useful when studying interactions between COPD-like lesions and tumorigenesis. A further toxin is cadmium chloride, a constituent of cigarette smoke. Administration of this substance also leads to alterations in pulmonary integrity with primarily interstitial fibrosis with tethering open of airspaces [ 147 ]. A combination of cadmium and lathyrogen beta-aminopropionile enhances emphysematous changes [ 148 ]. Tissue-degrading approaches Emphysema-like lesions can also be achieved by intrapulmonary challenge with tissue-degrading enzymes and other compounds [ 149 ] (Figure 2 ). Proteinases such as human neutrophil elastase, porcine pancreatic elastase, or papain produce an efficient enzymatic induction of panacinar emphysema after a single intrapulmonary challenge [ 150 , 151 ]. Since bacterial collagenases do not lead to the formation of emphysema, the effectiveness of the proteinases is related to their elastolytic activity. While these models may not be as useful as smoke exposure studies to achieve COPD-like lesions, they can lead to a dramatic picture of emphysema and may be used to study mechanisms related specifically to emphysema and to the repair of damaged lung. However, the method of inducing emphysema-like lesions by intratracheal instillation of these enzymes may not very closely relate to mechanisms found in the human situation. Among the different emphysema models, elastase-induced emphysema has also been characterized to be accompanied by pulmonary function abnormalities, hypoxemia, and secretory cell metaplasia which represent characteristic features of human COPD. Recent studies suggested that exogenous retinoic acid can induce alveolar regeneration in models of elastase-induced experimental emphysema [ 152 ] and that retinoic acid may have a role for alveolar development and regeneration after injury [ 153 , 154 ]. However, the role of retinoic acid in relation to alveolar development has only been analysed in a rat model and models in other animals did not show similar effects [ 155 ]. Also, the ability of alveolar regeneration which is present in rats does not occur to a similar extent in humans; a recent clinical trial using retinoic acid in COPD did not show positive results [ 156 ]. The mechanisms of emphysema induction by intratracheal administration of elastase encompass an initial loss of collagen and elastin. Later, glycosaminoglycan and elastin levels normalize again but collagen levels are enhanced. The extracellular matrix remains distorted in structure and diminished with resulting abnormal airway architecture [ 157 ]. The enlargement of the airspaces immediately develops after the induction of elastolytic injuries and is followed by inflammatory processes which lead to a transformation of airspace enlargement to emphysema-like lesions. This progression most likely occurs due to destructive effects exerted by host inflammatory proteinases. Addition of lathyrogen beta-aminopropionile leads to an impairment of collagen and elastin crosslinking and therefore further increases the extent of emphysema-like lesions [ 158 ]. Effects seem to be mediated via IL-1β and TNFα receptors since mice deficient in IL-1β Type1 receptor and in TNFalpha type 1 and 2 receptors are protected from developing emphysema following intratracheal challenge with porcine pancreatic elastase. This was associated with reduced inflammation and increased apoptosis [ 159 ]. In general, intrapulmonary administration of tissue-degrading enzymes represents a useful tool especially when focusing on mechanisms to repair emphysematic features. However, the lack of proximity to the human situation needs to be realized since the mechanisms of emphysema induction are clearly not related to the human situation. An advantage of proteinase-based models is the simple exposure protocol with a single intratracheal administration leading to significant and rapid changes. However, extrapolating these findings to slowly developing features of smoking induced human COPD is very difficult since a large number of mediators may not be involved in the rapid proteinase approach. Therefore, these models may not encompass important features of human COPD which may be more closely mimicked by inhalation exposures and it is clear that tissue-degrading enzyme models always represent the picture of an "induced pathogenesis". Gene-targeting approaches The genetic predisposition to environmental disease is an important area of research and a number of animal strains prone to develop COPD-like lesions have been characterized [ 160 - 162 ] (Figure 2 ). Also, genetically-altered monogenic and polygenic models to mimic COPD have been developed in recent years using modern techniques of molecular biology [ 163 , 164 ]. Gene-depletion and -overexpression in mice provide a powerful technique to identify the function and role of distinct genes in the regulation of pulmonary homeostasis in vivo . There are two major concepts consisting of gain-of-function and loss-of-function models. Gain-of-function is achieved by gene overexpression in transgenic mice either organ specific or non-specific while loss of function is achieved by targeted mutagenesis techniques [ 165 , 166 ]. These models can be of significant help for the identification of both physiological functions of distinct genes as well as mechanisms of diseases such as COPD. A large number of genetically-altered mice strains have been associated to features of COPD and a primary focus was the assessment of matrix-related genes. As destruction of alveolar elastic fibres is implicated in the pathogenic mechanism of emphysema and elastin is a major component of the extracellular matrix, mice lacking elastin were generated. It was shown that these animals have a developmental arrest development of terminal airway branches accompanied by fewer distal air sacs that are dilated with attenuated tissue septae. These emphysema-like alterations suggest that in addition to its role in the structure and function of the mature lung, elastin is essential for pulmonary development and is important for terminal airway branching [ 167 ]. Also, deficiency of the microfibrillar component fibulin-5 and platelet derived growth factor A (PDGF-A) leads to airspace enlargement [ 168 , 169 ]. PDGF-A(-/-) mice lack lung alveolar smooth muscle cells, exhibit reduced deposition of elastin fibres in the lung parenchyma, and develop lung emphysema due to a complete failure of alveogenesis [ 170 ]. The postnatal alveogenesis failure in PDGF-A(-/-) mice is most likely due to a prenatal block in the distal spreading of PDGF-R alpha+ cells along the tubular lung epithelium during the canalicular stage of lung development [ 170 ]. The importance of integrins in causing emphysema has been demonstrated in mouse. Epithelial restricted integrin α vβ 6-null mice develop age-related emphysema through the loss of activation of latent TGF-beta which leads to an increase in macrophage MMP-12 expression [ 171 ]. Fibroblast growth factors are known to be essential for lung development. Mice simultaneously lacking receptors for FGFR-3 and FGFR-4 have an impaired alveogenesis with increased collagen synthesis [ 172 ]. It is crucial to distinguish developmental airspace enlargement from adult emphysema which is defined as the destruction of mature alveoli. However, the identification of numerous factors influencing lung development is an important step towards identifying potential mechanisms underlying the development and progression of emphysema in human COPD. Next to developmental airspace enlargement also spontaneous emphysema may occur in genetically-modified mice strains and a gradual appearance of emphysema-like lesions has been found in mice lacking the surfactant protein D (SP-D) gene [ 173 ] and in mice lacking the tissue inhibitor of metalloproteinase-3 (TIMP-3) gene [ 174 ]. In these strains, matrix metalloproteinases were suggested to be the primary mediators of tissue destruction. A further mechanism to induce emphysema-like lesions is to expose developmentally normal genetically-modified animals to exogenous noxious stimuli such as tobacco smoke. This also allows identifying potential molecular mechanisms involved in the pathogenesis of COPD. Using macrophage elastase (MMP-12) gene-depletion studies it was shown that in contrast to wild type mice, the lung structure of MMP-12 gene-depleted animals remains normal after long term exposure to cigarette smoke [ 47 ]. These animals also fail to develop macrophage accumulation in response to cigarette smoke, an effect that could be related to MMP-12 induced generation of elastin fragments that are chemotactic for monocytes [ 175 , 176 ]. In summary, gene-targeting techniques display very useful tools to examine potential molecular mechanisms underlying human COPD. In combination with inhalation protocols they may identify important protective or pro-inflammatory mediators of the disease. Other models Various other agents have also been characterized to induce airway inflammation injury. In this respect, administration of toxins such as endotoxin leads to a recruitment of neutrophils and macrophage activation with concomitant airspace enlargement [ 177 , 178 ]. Non-inflammatory emphysema-like lesions may also be accomplished by intravascular administration of a vascular endothelial cell growth factor receptor-2 (VEGFR-2) blocker [ 179 ]. VEGF is required for blood vessel development and endothelial cell survival and its absence leads to endothelial cell apoptosis [ 180 ]. An increased septal cell death in human emphysematous lungs and a reduced expression of VEGF and VEGFR-2 is found in emphysema lungs [ 181 ]. Also, chronic blockage of VEGFR-2 causes alveolar septal cell apoptosis and airspace enlargement [ 179 ]. These findings of airspace enlargement point to a role of the vascular system in the development and progression of emphysema. Conclusions In contrast to the variable pathology and different stages of severity in human COPD, currently available animal models are restricted to mimicking a limited amount of characteristic features of COPD. Animal models need to be precisely evaluated based on whether they agree with features of human COPD in order to advance the understanding of mechanisms in human COPD. Based on inhalative exposure to noxious stimuli such as cigarette smoke, the administration of tissue-degrading enzymes or gene-targeting techniques, a number of experimental approaches to mimic acute and chronic features of COPD have been established in the past years. Due to the complexity of the disease, and species-specific differences they are all limited concerning their clinical significance. While the induction of the COPD lesions by tissue-degrading enzymes may appear artificial in many cases, it does not mean that these models are not valuable because they can be used to study many aspects of pulmonary pathophysiology of end-stage emphysema. Cellular mechanisms can be studied efficiently and underlying molecular mechanisms and potential therapeutic approaches can be revealed if the data is extrapolated cautiously. Combined models of inhalative exposure, proteinase-based tissue degradation to produce emphysema and gene-targeting techniques may provide models of COPD which encompass more features of the disease. However, one cannot assume that reproducing COPD with a high degree of fidelity in the animal necessarily means that the model simulates the human condition. In fact, a model that only produces a single pathologic COPD feature may be more useful as long as it produces this feature via a relevant mechanism that allows exploratory research. By contrast, a model producing all kinds of COPD features via irrelevant mechanisms may be less useful. In this respect, validation of models as being relevant is an extremely important issue in the early steps of model development. Animal models should not only assess histopathological features but also attempt to focus on functional features of human COPD such as airflow limitation, mucus hypersecretion, chronic cough and exacerbations, and also on pharmacological features such as corticosteroid resistance or diminished β-adrenergic bronchodilator responses. In conclusion, there are many benefits that can accrue from the development of animal models of COPD, most important of which is understanding of mechanisms and development of specific drugs for COPD.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC533858.xml
520826	Daily rhythms in plasma levels of homocysteine	Background There is accumulated evidence that plasma concentration of the sulfur-containing amino-acid homocysteine (Hcy) is a prognostic marker for cardiovascular morbidity and mortality. Both fasting levels of Hcy and post methionine loading levels are used as prognostic markers. The aim of the present study was to investigate the existence of a daily rhythm in plasma Hcy under strictly controlled nutritional and sleep-wake conditions. We also investigated if the time during which methionine loading is performed, i.e., morning or evening, had a different effect on the resultant plasma Hcy concentration. Methods Six healthy men aged 23–26 years participated in 4 experiments. In the first and second experiments, the daily rhythm in Hcy as well as in other amino acids was investigated under a normal or an inverse sleep-wake cycle. In the third and fourth, Hcy concentrations were investigated after a morning and evening methionine loading. To standardize food consumption in the first two experiments, subjects received every 3 hours 150 ml of specially designed low-protein liquid food (Ensure ® formula). Results In both the first and second experiments there was a significant daily rhythm in Hcy concentrations with a mid-day nadir and a nocturnal peak. Strikingly different 24-h patterns were observed in methionine, leucine, isoleucine and tyrosine. In all, the 24-h curves revealed a strong influence of both the sleep-wake cycle and the feeding schedule. Methionine loading resulted in increased plasma Hcy levels during both morning and evening experiments, which were not significantly different from each other. Conclusions There is a daily rhythm in plasma concentration of the amino acid Hcy, and this rhythm is independent of sleep-wake and food consumption. In view of the fact that increased Hcy concentrations may be associated with increased cardiovascular risks, these findings may have clinical implications for the health of rotating shift workers.	Background Experimental results accumulated in recent years have revealed that plasma concentration of the sulfur-containing amino-acid homocysteine (Hcy) is a prognostic marker for cardiovascular morbidity and mortality [ 1 - 5 ]. Plasma concentrations of Hcy in excess of 15 μmol/L under fasting conditions were associated with increased risk of cardiovascular mortality [ 6 ]. Furthermore, some patients having normal fasting levels of plasma Hcy were shown to have abnormally high levels of Hcy after methionine loading [ 7 ]. In most epidemiological studies, the differences between fasting concentrations of Hcy of cardiovascular patients and normal controls did not amount to more than 10–15%. Studies conducted during the 1960s have demonstrated that plasma levels of several amino acids vary in a daily manner. Feigin, Klainer and Beisel [ 8 ] were the first to report on daily rhythms in serum levels of total amino acids in adult men. The peak levels of the total integrated amino acids occured between 1200 and 2000 with a minimum level at 0400. Wurtman, Chou and Rose [ 9 ] reported on a daily rhythm in plasma concentration of tyrosine with a nocturnal nadir and a morning peak, which represented a two-fold increase in plasma tyrosine level. This rhythm persisted when subjects were maintained on a two-week low protein diet. Subsequently, the same group [ 10 ] extended their findings to 15 additional amino acids. Tyrosine, tryptophan, phenylalanine, methionine, cysteine, and isoleucine, underwent the greatest daily changes while alanine, glycine and glutamic acid showed the least. Hussein et al [ 11 ] reported that the daily fluctuations of plasma free amino acids were significantly affected by the dietary conditions. In none of these studies, however, were the levels of amino acids determined during the sleep period or under uniform dietary conditions. More recently, plasma Hcy levels were also shown to vary in a daily manner in humans with an evening peak and a morning nadir [ 12 ]. Significant daily rhythmicity was found in obese diabetic patients but not in normal controls. Since plasma samples were obtained every 3 hours and no attempt was made to examine how sleep affected the pattern of secretion, it is difficult to determine whether these findings bear any clinical significance. In rats, plasma Hcy demonstrated a 24-h rhythm with a nocturnal peak and a daytime nadir. Pinealectomy did not change the phase of the rhythm or its nocturnal elevation, but it did significantly increase mean plasma Hcy [ 13 ]. In the present study, we further investigated the possible existence of a daily rhythm in plasma Hcy under strictly controlled nutritional and sleep-wake conditions. We also investigated if the time during which methionine loading is performed, i.e., morning or evening, had a different effect on the resultant plasma Hcy concentration. Methods Subjects Six healthy men aged 23–26 years participated in 4 experiments. All were students who maintained a normal and regular sleep-wake cycle for at least three months prior to the studies. They were screened to ensure an adequate state of health by physical examination, detailed medical history and blood testing. All had a normal body weight (mean body mass index (BMI) = 23.5 ± 1.6 Kg/m 2 ). They were instructed to avoid alcohol and coffee beverages during the 24 hours that preceded each of the experimental periods. The study was approved by the local Human Ethics Committee, and subjects gave written informed conset before being enrolled in the first experiment. Subjects were paid for their participation. Procedure In the first and second experiments, daily rhythms in Hcy as well as in other amino acids were investigated under a normal or an inverse sleep-wake cycle. In the third and fourth experiments, Hcy concentrations were investigated after morning and evening methionine loading. Experiment 1 Subjcts were admitted to the laboratory at 1800 for a period of 24 hours, after having a normal day. A catheter was inserted into an antecubital vein and was kept patent by a drip of saline. Electrodes were attached for polysomnographic monitoring to determine sleep stages. These included EEG, EMG, EOG, respiration by respiratpry belt and nasal thermistor, and oximetry. Starting at 1900, 5-ml blood samples were drawn every hour until 1900 on the next day. Thoughout this period subjects were either in a supine or a sitting position in individual rooms where they could read, use their personal computers or watch television. From 2300 to 0800 the room lights were turned off during the sleep period. Blood samples were collected into EDTA treated tubes, immediately centrifuged at 4°C, and plasma was stored at -70°C until assay. Hourly blood sampling during sleep continued with minimal disturbance to subjects' sleep. To standardize food consumption and to provide adequate energy intake, subjects received every 3 hours 150 ml of specially designed liquid food (Ensure ® formula) with the following composition: proteins (5.49 g, 84% caseinate, 16% soy – 14.7% of the calories), fat (5.3 g, 32% of calories), carbohydrates (20 g, 77% corn syrup, 23% sucrose, 53.3% of calories), vitamins and minerals, in 77 ml water. No other food except for water was allowed. Experiment 2 Thes second experiment was identical to Experiment 1 except for the fact that the sleep period was delayed from 2300-0800 to 0720-1500. As before, subjects were admitted to the laboratory at 1800 and blood was withdrawn every hour starting at 1900 until 1900 on the next day. Sleep was monitored polygraphically as described before. Food was provided as in Experiment 1. Experiments 3 and 4 In these experiments, we conducted a methionine loading test at two times: 0900 and 2100. The selection of these times was based on the results of the first two experiments that demonstrated a daily nadir and a nocturnal peak in Hcy levels (see below). At the start of each methionine loading test, subjects were administered 100 mg/kg body weight methionine, mixed in fruit juice. Blood samples (5 ml) were taken into EDTA treated tubes before methionine loading, designated as time 0, and then at +2, +4, +6 and +8 hours after methionine administration. Light carbohydrate rich meals were provided at +1 and +6 hours after the methionine loading in each of the test periods. Measurement of amino acids and vitamins Plasma amino acids levels (Hcy, methionine, leucine, isoleucine and tyrosine) were measured in duplicates using a Biochrom 20 Amino-Acid analyzer (Pharmacia Biothech, Cambridge, UK) as described before [ 5 ]. The mean intra-assay CV was less than 3%. All samples from a single individual were analysed in a single run. In view of their involvment in Hcy metabolism, serum levels of folic acid and vitamin B 12 were also measured in all samples of all subjects using commercially available kits from Abbott. The assays were performed on an Abbott IMX analyzer that utilizes ion capture technology for folate determination and microparticle enzyme immunoassay (MEIA) technology for B 12 . The assays were performed according to the manufacturers' instructions and used quality control sera supplied by Abbott. Statistical analysis Repeated measurements ANOVA was used to compare the means of the amino acids between the first two experiments. To obtain the average 24-h Hcy curves, each individual data point was replaced by a z-transformation based on the individual 24-h mean and standard deviation, before averaging across subjects. Then, each of the individual time series was subjected to Cosinor analysis to determine its amplitude and acrophase. Since Experiment 1 was perfomed during the summer (August) and Experiment 2 was performed during early winter (late November), approximately 2 months after the change from Summer daylight-saving time to Winter time, during which the clock in Israel was advanced by one hour, the 24-h curves of the first experiment were advanced by 1 hour before the analysis. Then repeated measurements ANOVA was used to determine differences in acrophase between the experiments. In the third and fourth experimens, the concentrations of Hcy at times 0, 2, 4, 6, and 8 hours after methionine loading were analysed by repeated measures ANOVA to determine if there were any significant morning-evening differences in Hcy levels. Results All subjects successfully completed the four experiments. In experiment 1 when they slept from 2300 to 0700, average sleep latency was 22.2 ± 7.3 min, total sleep time was 407.3 ± 51.8 min, and sleep efficiency was 77.7 ± 9.2%. In experiment 2 when they slept from 0720 to 1500, average sleep latency was 4 ± 3.1 min, total sleep time was 371.5 ± 59.4 min, and sleep efficiency was 83.6 ± 12.2%. In spite of the reversal of the sleep-wake cycle, the 24 h means and coefficients of variation of Hcy in the two experiments were very similar to each other, 8.82 μmol/L and 29.7% and 8.51 μmol/L and 27.7%, in experiments 1 and 2, respectively. None of the subjects had abnormal Hcy levels (>15 μmol/L) at any point across the 24 hours. Figure 1 presents the average z-transformed 24-h curves of Hcy in the two experiments. In spite of the reversal of the sleep-wake cycle, the 24-h pattern of Hcy was remarkeably similar. In both experiments there was a midday nadir and a nocturnal peak in Hcy levels. In absolute terms, the daily rhythm in Hcy represents a change from nadir to peak values of 6.7 to 9.83 μmol/L (46.7%) and 7.4 to 10.55 μmol/L (42.6%), in experiments 1 and 2, respectively. Analysis of variance showed no significant difference in the average amplitude of the z-transformed rhythms of the two experiments, as determined by the cosinor analysis: 0.81 ± 0.19, and 1.07 ± 0.22 μmol/L, for experiment 1 and 2, respectively. There was, however, a significant difference between the timing of the average acrophase which was earlier by approximately 2 hours in experiment 1 than in experiment 2 (22:47 ± 0:45 vs. 0:54 ± 1:14, t = 3.77; p < .01). Figure 1 Daily rhythms in plasma concentration of Homocysteine. Rhythms were measured in 6 subjects who slept from 23:00 to 07:00 (Night sleep) or from 07:20 to 15:00 (Day sleep). Blood was withdrawn every hour starting at 19:00 until 19:00 the next day. Individual data points were transformed to Z-scores before averaging across subjects. For clarity purposes standard errors of data points are not presented. Magnitude of standard errors was approximatly 10% of mean values. Strikingly different 24-h patterns were observed for the other amino acids: methionine, leucine, isoleucine and tyrosine. In all, the average z-transformed 24-h curves revealed a strong influence of both the sleep-wake cycle and the feeding schedule. Their level was notably lower during the sleep period, regardless of its timing, and increased every two hours in synchrony with the times of feeding. This pattern is exemplified in Figure 2 for methionine. Identical patterns were observed for leucine, isoleucine and tyrosine (data not shown). Figure 2 Daily rhythms in plasma concentration of methionine. Rhythms were measured in 6 subjects who slept from 23:00 to 07:00 (Night sleep) or from 0720 to 1500 (Day sleep). Blood was withdrawn every hour starting at 19:00 until 19:00 the next day. Individual data points were transformed to Z scores before averaging across subjects. For clarity purposes standard errors of data points are not presented. Magnitude of standard errors was approximatly 10% of mean values. Note the large pulses in methionine concentrations that appeared in synchrony with the times of feeding. We did not find any evidence for rhythmicity in the concentrations of B 12 and folic acid. While folic acid showed a linear increase throughout the study period, the 24-h pattern of B 12 was rather constant with slight elevation during the night time (data not shown). Methionine loading As expected, methionine loading resulted in increased plasma Hcy levels during both morning and evening experiments (Figure 3 ). Analysis of variance did not reveal overall significant differences between morning and evening post-methionine Hcy levels. However, inspection of Hcy levels at each of the time points separately revealed some interesting trends. Before methionine loading, as could be expected from the daily rhythm in Hcy found in experiments 1 and 2, morning Hcy level tended to be lower by 1.18 μmol/L than the evening level (p < .11, paired t-test, two tailed). Moreover, the increase in Hcy from time 0 to 2 hours after loading was greater by a mean of 2.8 μmol/L in the evening than in the morning (p < .09, paired t-test, two tailed). This resulted in evening and morning levels of Hcy of 26.66 and 23.86 μmol/L, respectively. These differences became much smaller at +4, +6 and +8 after the loading. Figure 3 Plasma concentration of homocysteine before and after methionine loading. Shown are the means and standard deviations of plasma concentration of homocysteine in 6 subjects before (0 hr) and 2, 4, 6 and 8 hours after methionine loading at 09:00 and 21:00. Discussion The present study demonstrated that under strictly controlled dietary conditions plasma levels of Hcy shows significant daily rhythmicity, which is independent of the 24-h cycle of sleep and wake, with a peak at around 2200 to 2400. Previously, similar rhythmicity in Hcy with an evening peak was reported in obese diabetic patients by Bremner et al [ 12 ] and with nocturnal peak in rats by Baydas et al [ 13 ]. We further extended these findings by demonstrating that daily rhythms exist also in normal young adults. In contrast to Hcy, there was no daily rhythmicity in methionine, leucine, isoleucine and tyrosine, in which the 24-h pattern followed both the timing of sleep and the feeding schedule. Homocysteine is a non-protein sulfur containing amino acid, and an intermediate in the metabolism of the essential amino acid methionine. The metabolism of Hcy is accomplished by two major pathways, remethylation into methionine and transsulfuration to cystationine [ 14 ]. In remethylation, Hcy acquires a methyl group from N-5-methyltetrahydrofolate or from betaine to form methionine. The reaction with N-5-methyltetrahydrofolate is vitamin B 12 dependent while the reaction with betaine is not. In the transsulforation pathway, Hcy condenses with serine to form cystationine in an irreversible reaction catalyzed by the pyridoxal-5'-phosphate (PLP)-containing enzyme, cystationine beta synthase. Although we do not have any information as yet on the underlying mechanism responsible for the daily rhythm in plasma Hcy, it is most probably related to the balance between its rates of production and disposal. A high Hcy concentration could be due to an elevated production rate, a decreased rate of transsulforation, a decreased rate of remethylation to methionine, or any combination of these processes. The fact that the range of the daily variations in the plasma levels of Hcy is on the same order of magnitude as those seen in mild hyperhomocysteinemia, may suggest that the two phenomena share a common underlying mechanism. Mild hyperhomocystenemia seen under fasting conditions is due to mild impairement in the methylation pathway. This may be caused by folate or B 12 deficiencies, or by methylenetetrahydrofolate reductase thermolability. The variations in plasma vitamin concentrations, however, could not provide an explanation for the daily rhythms in Hcy. The 24-pattern of folate levels showed a linear increase from the beginning to the end of the study. Although the plasma concentrations of vitamin B 12 varied across the 24 hours – in contrast however to what was expected if B 12 were involved in the daily rhythm in Hcy, ie, increasing levels of B 12 associated with decreasing levels of Hcy – the 24-h pattern in B 12 was parallel to that of Hcy with a daytime nadir and a night time peak. Thus, it is unlikely that a daily rhythm in plasma vitamin concentrations can explain the daily rhythm in Hcy. The methionine loading test has been used to test the individual's ability to dispose of methionine through the transsulforation pathway [ 14 ]. The fact that the differences between Hcy levels after morning and evening methionine loading were rather small and limited to the first 2 hours after the loading may indicate that the transsulforation pathway does not play a role in generating Hcy rhythmicity. A different possibility that cannot be ruled out at this point is the involvement of the Hcy cellular export mechanism. The small amount of plasma Hcy is the result of a cellular export mechanism that is essential for keeping intracellular concentrations low to avoid potentially Hcy cytotoxic effects. Thus the daily rhythm in plasma Hcy may reflect variations in the activity of the cellular export mechanism, which result in varying levels of Hcy disposed to the plasma at different phases of the 24 hours rather than in its rate of metabolism. Further studies are needed to test this possibility. Finally, what may be the clinical implications of the present findings? We would like to suggest that the existence of a daily rhythm in Hcy concentration may have possible health-related consequences to shift workers, who were shown to be at an increased cardiovascular risk [ 15 ]. Firstly, reversing the meals' schedule to a nocturnal orientation such that the time of major meal coincides with the time of the physiological peak of Hcy may have at least transient cardiovascular consequences. It was shown that an increase in Hcy concentration rapidly induces impaired elasticity of the coronary microvascular and central arterial circulation [ 16 , 17 ], conditions predictive of increased cardiovascular events rate [ 18 ]. Furthermore, even small physiological increments in Hcy concentration, induced by low-dose methionine or dietary animal protein meals that are more relevant to shift workers, induce a dose-related graded impairement in endothelial functioning [ 19 ]. Thus, consuming methionine or animal-protein-rich foods during the middle of the night may result in a greater risk of severe transient impairment in endothelial function than when a similar meal is consumed at the habitual lunch time during the day. Although we did not find significant differences in Hcy concentrations after methioning loading at 0900 and 2100, as expected, morning levels tended to be lower, and the initial increase in Hcy during the first 2 hours after loading was greater by a mean of 2.8 μmol/L in the evening than in the morning. This difference bordered on statistical significance. It is possible that, had we performed the methinine loading closer to the time of the nocturnal peak in Hcy, between 10 PM and midnight, this day-night difference would have been larger. Secondly, we do not know how the desynchronization between the circadian system and the enviornment which occurs in rotating shift workers may affect the rhythm in Hcy concentrations and its overall plasma concentration. Recently, Martins et al [ 20 ] reported that long-haul bus drivers working shifts had higher concentrations of Hcy than a control group of day workers. In a study just completed in our laboratory we found that rotating shift workers who complained of disturbed sleep had significantly higher concentrations of Hcy than permanent day workers, or shift workers without sleep disturbances (paper submitted to press). Furthermore, life-style related factors like smoking and heavy coffee consumption that were shown to be associated with increased Hcy concentration [ 21 , 22 ], are more prevalent among shift workers than among day workers [ 23 ], and may also contribute to increased Hcy concentration. Of note, decreasing levels of melatonin induced by pinealectomy in rats were reported to be associated with increased plasma concentrations of Hcy, while treatment with exogenous melatonin restored it to basal concentrations [ 24 ]. Thus, suppression of melatonin by bright light during night work may be also associated with increased Hcy concentration. In view of the fact that Hcy is a risk factor for cardiovascular morbidity, more research is needed on the possible role of hyperhomocysteinemia as a cardiovascular risk factor in shift workers. Conclusions Our results demonstrated a daily rhythm in plasma concentrations of Hcy with a nocturnal peak that was independent of sleep-wake cycle and food consumption. There were no comparable rhythms in the concentrations of methionine, leucine, isoleucine and tyrosine, nor in the concentrations of B 12 and folic acid. Methionine loading at 9 AM and 9 PM produced a comparable time-dependent increase in Hcy concentrations with a tendency toward a higher increase in the evening during the first 2 hours after loading. In view of the possible involvement of Hcy in cardiovascular morbidity, and of the increased cardiovascular morbidity in shift wokers, these findings may have implications to shift workers health. List of abbreviations Hcy – homocysteine EEG – Electroencephalography EMG – electromyography EOG – electrooculography EDTA – ethylanediaminetetraacetic acid CV – coefficient of variation ANOVA – analysis of variance Competing interests None declared. Author's contribution PL and LL co-designed the study, supervised the data collection and data analysis and wrote the paper.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC520826.xml
534106	Healthcare professionals' perceptions of pain in infants at risk for neurological impairment	Background To determine whether healthcare professionals perceive the pain of infants differently due to their understanding of that infant's level of risk for neurological impairment. Method Neonatal Intensive Care Units (NICU's) at two tertiary pediatric centers. Ninety-five healthcare professionals who practice in the NICU (50 nurses, 19 physicians, 17 respiratory therapists, 9 other) participated. They rated the pain (0–10 scale and 0–6 Faces Pain Scale), distress (0–10), effectiveness of cuddling to relieve pain (0–10) and time to calm without intervention (seconds) for nine video clips of neonates receiving a heel stick. Prior to each rating, they were provided with descriptions that suggested the infant had mild, moderate or severe risk for neurological impairment. Ratings were examined as a function of the level of risk described. Results Professionals' ratings of pain, distress, and time to calm did not vary significantly with level of risk, but ratings of the effectiveness of cuddling were significantly lower as risk increased [ F (2,93) = 4.4, p = .02]. No differences in ratings were found due to participants' age, gender or site of study. Physicians' ratings were significantly lower than nurses' across ratings. Conclusion Professionals provided with visual information regarding an infants' pain during a procedure did not display the belief that infants' level of risk for neurological impairment affected their pain experience. Professionals' estimates of the effectiveness of a nonpharmacological intervention did differ due to level of risk.	Background Research on pain in infants has progressed considerably over the past twenty years. The nature and frequency of procedural pain in the neonatal intensive care unit (NICU) is now understood [ 1 ], many measures have been developed for assessment of acute pain in the NICU [ 2 ] and many pain interventions have now been evaluated [ 3 ]. However, much less is known about the pain experienced by neonates who are at risk for neurological impairment (NI), as most studies of neonatal pain have either excluded this group or have not examined data specific to them within larger data sets. We do know that this group represents approximately 10% of infants admitted to the Neonatal Intensive Care Unit [ 4 ] and that they experience more painful procedures in the NICU during the first days of life than infants who are not at risk for NI [ 5 ]. It also appears this vulnerable group may be particularly susceptible to potential long-term negative consequences of pain because of their neurological fragility, concomitant illnesses, and repeated exposure to painful stimuli [ 6 ]. The crucial first step of pain management is pain assessment. Without a valid and reliable approach to assessing pain, and the demonstrated efficacy of interventions for pain, decisions about pain management may not improve care. However, even with valid, reliable pain assessment tools, the characteristics of healthcare providers may affect ratings provided by them. These characteristics can include the healthcare providers' views of pain interventions [ 7 ], lack of awareness of advances in pain management [ 8 ], and use of pain cues that are not reliable [ 9 ]. The present study was designed to move beyond self-report of beliefs to examine whether healthcare professionals' judgments of pain in neonates are affected by their perception that a neonate has mild, moderate or severe risk for neurological impairment. Research in this area is only emerging, but has important implications for how healthcare professionals deliver care to this vulnerable population. In a previous study using questionnaires, we found that caregivers of children with severe cognitive impairment view the pain of children with more severe impairment as reduced [ 10 ]. A second study using the same questionnaire with healthcare professionals and students revealed a similar pattern of beliefs [ 11 ]. Most recently, we adapted that questionnaire to assess the beliefs of healthcare professionals' regarding the pain of infants with varying degrees of risk for neurological impairment and again found that those who took part believed that the degree of pain experienced decreases as risk for neurological impairment increases [ 12 ]. These studies suggest that those who manage the pain of infants at risk for, or children with, intellectual deficits believe that pain is less for those at greater risk or who have greater impairment. These results may explain why we also found infants at risk for neurological impairment receive less pain treatment in the NICU [ 13 ]. However, to extrapolate from questionnaires to clinical behaviour can be problematic. Thus, the current study was designed as a step to linking these two sets of results. Specifically, we felt it was important to know if professionals' beliefs about pain in this group influence their assessment of infants' pain, which could lead to those infants' being provided with less pain treatment. As with any experimental study, the circumstances could not completely replicate those in a clinical setting. For example, the participants would not have access to physiological data or to the infants' recent history of pain. However, we hypothesized that the participants in this study would rate the pain lower for infants described as having greater risk for neurological impairment, corresponding with the beliefs expressed in our three previous studies. Methods Participants One hundred and one healthcare professionals, with at least one year of experience working with infants with neurological impairment in the NICU, were recruited from two tertiary level university affiliated NICU settings in central and eastern Canada. They were recruited through information provided by the research nurses and notices posted within their centers. Each participant was paid a small honorarium for their participation, and all provided informed consent. The study was approved by each health centre's Research Ethics Board. Materials Demographic information Participants completed a questionnaire that requested information regarding their age, gender, education and work experience. Video clips Nine video clips were viewed by each participant. The 30 second video clips depicted term and preterm neonates of Caucasian descent experiencing a heel stick and squeezing for blood collection. Video clips were of the infants' faces only, with most lying on their sides and all bundled. Audio was not included. Prior to each videoclip, a verbal description of the neonate suggesting he/she was at mild, moderate or severe risk for neurological impairment was provided to the participant. These descriptions had been previously recorded on audio tape by a researcher to ensure each participant was read the description in an identical fashion. Descriptions were counterbalanced such that each videoclip was described for some participants as having either mild risk, moderate risk, or severe risk within each of two orders of presentation. Thus, each participant viewed three infants that were described as having mild, moderate or severe risk for neurological impairment, but the level of risk, and the order in which the clips were presented varied. Examples of the descriptions provided to participants are shown in Table 1 . Table 1 Sample descriptions of infants viewed on videotape provided Mild Risk Moderate Risk Severe Risk Brianna is 6 days old and has been treated for neonatal jaundice. She will make a complete recovery. Otherwise she is healthy. Samuel was born 4 weeks prematurely and was mildly asphyxiated at birth because the cord was wrapped around his neck during delivery. An MRI shows a small area of damage in the brain. Matthew was born with a serious metabolic condition which caused significant brain damage. He will likely not survive past 2 years of age. Jason was born prematurely and is gaining weight slowly. He is now one month old. He suffered a unilateral Grade I bleed in his brain and will likely have no permanent damage from that. Matthew was born with a serious metabolic condition which caused moderate brain damage. With the aid of a special diet, he will develop fairly well but will likely have significant learning disabilities. Samuel was born 4 weeks prematurely and was severely asphyxiated at birth because the cord was wrapped tightly around his neck at delivery. An MRI shows extensive damage throughout the brain. Ratings of pain and distress Participants rated the pain of infants shown on videotape from 0 (no pain) to 10 (extreme pain). They also rated each infant's pain using a Faces Pain Scale (0 = no pain, 6 = extreme pain). These measures were chosen because they are easy to use and were feasible for this experimental task. Although many validated measures of neonatal pain exist, these are multidimensional in nature. As such, they require the person using them to have access to information regarding the infant's physiological status, something we were unable to provide in the context of this task. The 7 face scale [ 14 ], was included to allow a check of the validity of the 0–10 pain rating, since the latter is not commonly used in clinical neonatal settings. The Faces Pain Scale is also not typically used in a neonatal setting, but research indicates most adults find it easy to use [ 15 ], making it a useful check of participants' 0–10 pain ratings. Preliminary analyses indicated there was a significant relationship, similar to results for reliability computed for other sets of observational pain tools used in pediatric research [ 16 ], between 0–10 pain ratings and Faces Pain Scale ratings for infants at mild ( r = .74), moderate (. r = .56) and severe risk for neurological impairment ( r = .60). Ratings of cuddling and calm Participants provided a rating of how effective they believed a behavioural intervention (i.e. cuddling) would be for minimizing the procedural pain for each infant (0 = no effect, 10 = very effective) and of how long (seconds) they believed it would take each infant to calm without intervention. Ratings of risk for neurological impairment To ensure that the descriptions provided were valid depictions of infants at each level of risk and were understood and accepted by participants, participants were asked to rate the level of risk they believed each infant had for future neurological impairment (0 = no risk, 10 = certain impairment). Procedure Participants took part in small groups of 5 to 6 professionals that were randomly assigned to one of the two orders of presentation. They completed the demographic questionnaire and the rating tasks were explained to them. They were then shown the nine video clips. After each videoclip was viewed, the participants were provided with time to make their independent ratings for that videoclip before the next was shown. Participants were not permitted to interact with each other until all tapes had been viewed and rated. After ratings were complete, they were debriefed regarding the purpose of the study and a discussion of their experience was facilitated by the research assistant. Preliminary analyses Exclusions due to missing data Preliminary analyses indicated that two participants were missing more than 10% of ratings. As per an a priori decision as to how to handle missing data, their data were excluded from further analyses. The remaining 99 participants were missing 0% ( n = 88) to 7% ( n = 1) of responses ( M = 0.2, SD = 0.6). Exclusions due to presentation order effects A 2 (group) X 6 (rating) Repeated Measures Analysis of Variance (RM ANOVA) was conducted on the six ratings provided by participants who viewed the tapes in the two orders to determine whether order of presentation had affected ratings. This analysis revealed a significant effect of order of presentation [ F (1,97) = 4.4, p = .04). A more detailed look at the data using stem and leaf plots revealed 4 participants in one group had extreme scores for ratings of Time to Calm ( M = 192.8, SD = 39.2) relative to the other participants in that group ( M = 37.4, SD = 28.7). The data of these four participants were removed. A second RM ANOVA revealed no significant effect due order of presentation of the video clips [ F (1,93) = 1.7, p = .20). Thus, 95 professionals formed the final sample for the study. Manipulation check To determine if the descriptions provided to the participants were effective in leading them to believe the infants viewed had mild, moderate or severe risk for neurological impairment, each participant's mean rating for degree of risk for future impairment for the infants they were told had a mild (3), moderate (3) or severe risk (3) for impairment were computed. These ratings were then compared using a RM ANOVA. There was a significant difference in the ratings provided to those clips described as having mild ( M = 3.8, SD = 2.8), moderate ( M = 4.9, SD = 2.0) and severe risk ( M = 6.1, SD = 2.8; F (2, 93) = 13.9, p < .001). Post-hoc paired t-tests indicated these differences were significant between infants described as having mild and moderate risk ( t ((94) = -3.2, p = .002), mild and severe risk ( t ((94) = -4.8, p < .001), and moderate and severe risk ( t ((94) = -5.0, p < .001), Thus, participants believed the infants had different levels of risk when they provided ratings for the video clips. Statistical procedures The data were analyzed using SPSS Version 10.0.7 [ 17 ]. Power computations were completed using Sample Power 1.2 [ 18 ] or based on tables prepared by Stevens [ 19 ]. Alpha was set at .05 for all tests and Bonferroni corrections were applied to sets of post hoc matched sample t-tests to maintain alpha at .05 for each set. Because the corrected p values varied with the number of tests in each set, raw p values are reported. Wilks Lambda was used to test significance for all RM ANOVA's. There was .80 power or greater to detect medium size effects using repeated measures analyses with 3 to 5 levels of factors and greater than .99 power to detect medium size differences in means using matched sample t-tests. Power was .86 to detect a significant medium size correlation between ratings and years of experience. Descriptive statistics Descriptive statistics were generated for the demographic characteristics of the participants (Table 2 ) and for the ratings provided for each set of video clips (Table 3 ). Table 2 Characteristics of the participants (N = 95) Characteristic n % Site Toronto 46 48 Halifax 49 52 Profession Nurse 50 53 Physician 19 20 Respiratory Therapist 17 18 Occupational Therapist 2 2 Physiotherapist 2 2 Psychologist 1 1 Other Clinician 4 4 Gender Female 82 86 Age 20–30 years 19 20 31–35 years 15 16 36–40 years 32 34 41–45 years 11 12 46 years or more 18 19 Note. Percentages rounded. Table 3 Mean ratings given to infants described as at risk for mild, moderate or severe risk for neurological impairment (N = 95) Rating Mild Risk Moderate Risk Severe Risk M SD M SD M SD Pain (0–10) 6.3 1.7 6.6 1.7 6.2 1.6 Faces pain Scale (0–6) 4.7 1.0 4.8 0.8 4.8 1.0 Distress (0–10) 6.1 1.7 6.5 1.5 6.3 1.6 Effectiveness of Cuddling (0–10) 7.0 2.1 7.1 1.9 6.5 2.2 Time to calm (Seconds) 35.2 34.8 34.9 28.3 32.8 31.8 Effect of risk for neurological impairment on ratings To compare the ratings provided for the 9 video clips, a 3 (level of risk) X 5 (rating type) RM ANOVA was conducted on the scores of the 95 participants. This was followed by 5 one-way RM ANOVA's on each rating (0–10 pain rating, Faces Pain Scale rating, distress rating, effectiveness of cuddling rating, time to calm) and matched sample t-tests on ratings when the one-way ANOVA was significant. Effect of participants' characteristics on ratings The effect of participants' characteristics on ratings was examined using Mixed Measures ANOVA's on the five ratings at three levels of risk. The first three included Gender, Age, and Site (Toronto, Halifax) as between-subjects effects. The fourth included three levels of profession (i.e. staff nurse, physician, respiratory therapist) as the between-subjects effect. Other professionals were not included due to small numbers. The relation between the participants' years of experience in a neonatal setting and their ratings were investigated using Pearson Correlations. Results Participants The characteristics of the participants are displayed in Table 2 . The majority were nurses and the number of years experience in a neonatal setting ranged from 1.5 to 36 years ( M = 11.8, SD = 7.7). The 50 nurses included staff nurses ( n = 34), advanced practice nurses ( n = 9) and nurse managers/educators ( n = 7). The physicians specialized in neonatology ( n = 10), neurology ( n = 4), pediatrics ( n = 3) and other specialties ( n = 2). Six of the 19 physician participants were residents or fellows. Additional professions are listed in Table 2 . Effect of risk for neurological impairment on ratings The mean ratings provided for video clips of infants described as having mild, moderate or severe risk for neurological impairment are depicted in Table 3 . The RM ANOVA on the five ratings revealed a nonsignificant effect of Level of Risk [ F (2,93) = 0.6, p > .05], a significant effect of Rating [ F (4,91) = 91.6, p < .001] and a significant interaction between the two [ F (8,87) = 3.4 p = .002]. Thus, there was no overall effect due to the level of risk described, but level of risk described did affect some ratings. One-way RM ANOVA's revealed Level of Risk had a marginal effect on participants' ratings of pain on the 0–10 scale [ F (2,93) = 2.9, p = .06] and a significant effect on ratings of the perceived effectiveness of cuddling [ F (2,93) = 4.4, p = .02], but nonsignificant effects on Faces Pain Scale ratings [ F (2,93) = 0.3, p = .70], distress [ F (2,93) = 2.2, p = .12] and time to calm [ F (2,93) = 0.4, p = .65]. As Table 3 shows, there was a slight tendency for participants to rate pain lower for infants who were described as having greater risk for impairment. Participants did believe cuddling would be less effective when risk for neurological impairment was greater. Ratings of the effectiveness of cuddling were significantly lower for infants described as at high risk than they were for those described as at mild risk [ t (94) = 2.5, p = .01] or moderate risk [ t (94) = 3.0, p = .004]. The difference in ratings between those described as at mild or moderate risk were nonsignificant [ t (95) = -0.2, p = .77]. Thus, participants believed that beyond a moderate level of risk, the effectiveness of cuddling dropped significantly. In summary, participants did not view the pain of the infants as varying due to level of risk for neurological impairment. Nor did they perceive the distress or time to calm after pain as differing between groups of infants described as having mild, moderate or severe risk for neurological impairment. However, they did perceive that cuddling would be less effective as an intervention for infants with high risk, than for those with mild or moderate risk of neurological impairment. Effect of participants' characteristics Site, gender and age Three Mixed Measures ANOVA's were used to examine the effect of participants' characteristics on the five ratings ratings. The first result indicates a nonsignificant main effect of Site [ F (1,93) = 0.7, p = .39], the second revealed a nonsignificant main effect of Gender [ F (1,93) = 0.9, p = .34], and the third indicated Age also did not significantly effect ratings on the five measures [ F (4,90) = 0.2, p = .95]. Thus, participants' ratings did not vary due to their institution, gender or age. Profession To examine the effect of participants' profession on their ratings, three groups were included in a Mixed Measures ANOVA: staff nurses (n = 34), physicians (n = 11), and respiratory therapists (n = 17). Residents and Fellows, Nurse Managers and Educators and Specialists, and other professionals were not included due to small numbers in those groups. The analysis revealed a significant main effect of Rating Scale [ F (4,56) = 3.9, p = .001]. However, the main effect of Level of Risk was nonsignificant and the main effect of Profession only approached significance [ F (2,59) = 2.7, p = .07]. Participants' ratings were not affected by their professional background. The interaction between Rating and Level of Risk was significant [ F (8,52) = 36.6, p < .001], but all other interactions were nonsignificant. Games-Howell post-hoc comparisons revealed a significant difference in the ratings provided by staff nurses and physicians ( p = .004) and a difference between respiratory therapists and physicians that approached significance ( p = .06). As shown in Table 4 , Nurses' ratings did not appear to differ greatly due to level of risk for impairment, while Physicians' showed a tendency to rate all aspects of the experience higher as level of risk increased, and respiratory therapists tended to provide lower ratings as the infants' level of described risk for neurological impairment increased. Table 4 Mean pain, distress, effectiveness of cuddling and time to calm scores given to infants described as at risk for mild, moderate or severe risk for neurological impairment by physicians and other clinicians Rating Level of Risk Staff Nurses (n = 34) Physicians (n = 11) Respiratory Therapists (n = 17) M SD M SD M SD 0 – 10 Pain rating Mild 6.4 1.6 4.8 2.3 6.9 1.3 Moderate 6.9 1.5 5.6 2.4 6.9 1.8 Severe 6.4 1.7 5.8 2.1 5.8 1.6 Faces pain rating (0–6) Mild 4.6 1.0 3.7 0.8 5.3 0.6 Moderate 4.9 0.7 4.8 0.6 4.9 1.1 Severe 4.9 1.0 5.4 0.6 4.3 1.3 0 – 10 Distress rating Mild 6.4 1.6 4.3 2.1 6.9 1.3 Moderate 6.8 1.3 5.6 2.2 6.8 1.9 Severe 6.4 1.6 6.3 1.9 5.4 1.6 0–10 Effectiveness of cuddling rating Mild 6.4 1.6 4.3 2.1 6.9 1.3 Moderate 7.2 2.0 6.9 1.9 6.5 2.4 Severe 6.6 2.3 5.8 2.6 4.9 2.1 Time to calm estimate (seconds) Mild 40.0 36.1 14.9 15.5 46.8 43.8 Moderate 39.8 28.9 18.4 13.8 42.2 35.4 Severe 40.5 40.3 23.3 15.7 31.3 33.0 Professional experience Eighty-nine participants provided information regarding their amount of professional experience. Correlations indicated that years of experience were not correlated significantly with any of the five ratings provided after corrections for multiple tests. Thus, the importance of an infants' level of risk for neurological impairment was neither greater nor less as experience in this setting increased. Discussion Overall, the professionals in this study did not rate the pain of neonates differently when provided with information indicating those infants had mild, moderate or severe risk for neurological impairment. The professionals' perception of the infants' level of risk also did not affect their ratings of the infants' distress, or their belief in how long the infant would take to calm after pain without intervention. Professionals did perceive that cuddling would be significantly less effective for infants at high risk for neurological impairment than for infants with mild or moderate impairment. However, this effect was not large, and, although it was statistically significant, it may be spurious. Further research should examine whether beliefs regarding pain experience in this group and beliefs regarding the effectiveness of cuddling and other nonpharmacological interventions are truly independent. These results are inconsistent with the results of our previous questionnaire study indicating professionals, with similar levels of experience in neonatal intensive care settings, perceive the pain experience of infants as reduced as their level of risk for neurological impairment increases [ 12 ]. There are several possible reasons for these discrepant results. The professionals who participated in this study were asked to rate the risk for neurological impairment of each infant they viewed on videotape. Asking them to do this may have alerted them to the purpose of the study and elicited efforts on their behalf to provide ratings that were unbiased. However, their ratings of the perceived effectiveness of cuddling did vary by level of risk for impairment, suggesting attempts to appear unbiased do not fully explain the results found. In our previous studies, questionnaires elicited beliefs about the pain experience of infants and children with varying levels of risk relative to the pain experience of those without risk [ 10 - 12 ]. In contrast, no infants in the current study were described as having no risk for neurological impairment. This was because the infants' appearance made it apparent that they were not healthy full-term infants. It may be that the comparative nature of the questions in the previous studies made the possibility of differences in pain experience due to neurological risk more salient to participants. Thus, the pain ratings provided here did not differ among levels of risk, but had ratings of healthy infants been included in the task, they may have differed significantly from them. It is also possible that the beliefs expressed by professionals in our previous study [ 12 ] do not moderate professionals' behaviour in relation to pain assessment for specific infants, as was found here. A discordance between expressed beliefs and behaviour, in regard to pediatric pain management, has been reported elsewhere [ 20 , 21 ]. Thus, the professionals here may hold similar beliefs to the professionals in our previous study, but these beliefs did not alter their behaviour when asked to judge pain in a specific infant based on observable behaviour. This interpretation is supported by the current results because no differences were found due to level of risk for ratings that the professionals could base on behaviour they observed on the video clips: pain, distress, time to calm. In contrast, professionals' judgments of the effectiveness of cuddling were influenced by the descriptions of the infants' level of risk for neurological impairment. This may be because there was no visual information to base this rating upon, so professionals used the descriptions of risk provided, presumably in light of their previous experience with these groups in the neonatal setting. The finding that pain ratings did not vary due to level of risk for neurological impairment raises questions about our previous study that revealed infants at risk for neurological impairment receive less pain treatment in the NICU [ 13 ]. When a group is provided less medication for pain, it is typically assumed that this is because their pain was judged as less. However, it is possible that professionals hold beliefs about pain treatment that directly impact upon treatment decisions, irregardless of pain assessment. For example, they may hold beliefs about the appropriateness of medication for specific groups that are unrelated to beliefs about the amount of pain that group experiences. In support of this perspective, research indicates that nurses hold negative attitudes towards pharmacological treatment for pain [ 7 ] and that steps to improve pain assessment do not necessarily result in changes in pain management [ 22 ]. Further research is needed to reconcile the current results with beliefs that risk for neurological impairment does affect pain experience expressed by a similar group of professionals in our previous survey [ 12 ] and the results of our study indicating procedural pain is not treated as frequently for infants in the NICU who have greater risk for neurological impairment [ 13 ]. If this reflects a disconnect between pain beliefs related to assessment and those related to treatment for infants at risk for neurological impairment, then educational interventions aimed at improving care through changes in pain assessment may be ineffective. In that case, other avenues to changing professionals' pain management for this group should be explored. Another finding in this study warrants discussion. Professionals' judgments of the effectiveness of cuddling decreased with increasing risk for neurological impairment, despite their having judged pain as similar in intensity. This result is similar to a finding by Fanurik et al. [ 23 ]. They found nurses, but not physicians, responding to vignettes of children undergoing painful procedures, indicated nonpharmacological interventions would be less appropriate as level of cognitive impairment increased. The same professionals' ratings of the pain intensity experienced by the children in that study did not differ due to perceived level of cognitive impairment. The current results, along with those of Fanurik's group [ 23 ], raise the question of whether professionals perceive the pain experienced by those at risk for or with neurological impairment as similar in intensity, but differing in quality from those at lesser risk. Because the current study elicited ratings only of the intensity of pain and distress and professionals were not asked about the nature of the pain the infants experienced, the results cannot confirm this possible explanation, as data regarding pain quality was not collected. However, professionals in our survey study differentiated between physiological aspects of pain and internal and external responses to pain, such as emotional reaction, behavioural reaction and communication of pain [ 12 ]. They also believed the experience of infants at greater risk was more reduced along the latter aspects that are more psychological in nature. Caregivers' have expressed similar beliefs, and also perceived the behaviour of children with more severe impairment is more closely related to their physiological pain experience [ 10 ]. From this finding, we could suggest that there is a belief, on the part of professionals and caregivers, that the pain behaviour of those at greater risk for, or with, neurological impairment is more reflexive in nature. We could further speculate that the underlying rationale may be that they are seen as less able to interpret their pain, both cognitively and emotionally, due to their neurological impairment. However, we would need to conduct further research to substantiate this rationale. If professionals and caregivers do believe pain behaviour is more reflexive, and that pain experience is more physiologically based when a child has neurological impairment, it could explain the current results regarding the effectiveness of cuddling. Professionals viewing the video clips may have perceived the behavioural responses of the infants with different levels of risk for impairment as being similar in intensity. Nonetheless, they may have interpreted the behaviour of those with more risk as more of a reflexive response to a physiological insult, while they saw the behaviour of those with lesser risk as reflecting a more multidimensional pain experience incorporating both physical and psychological suffering. Thus, we could again speculate that they may have felt cuddling, an intervention that would address physical and psychological aspects of pain, would be more effective for the less impaired groups. This phenomenon would not be novel or unique. For most of recorded history, there has been a belief that cognitive interpretation of pain was necessary for pain to result in long-term negative consequences. This belief was often the justification for poorer pain management for both children and infants [ 24 ]. Although this belief is fading in regard to children and infants in general, it is still held in relation to those who are most severely at risk for, or have neurological impairment, and are perceived as least capable of interpreting their pain. Alternatively, this belief may be based on the actual experience of professionals in this study, that it is more difficult to calm an infant at risk for neurological impairment. This experience may also be an accurate perception of the difficulty infants at greater risk for impairment may have in responding to behavioural interventions because of their reduced ability to organize behavioural state and biobehavioural responses. Further research should examine these areas of speculation to specifically determine whether the perception that a behavioural intervention will be less effective for infants at greater risk for neurological impairment does reflect professionals' direct experience with this group or their understanding of how the pain experience may be affected by neurological impairment that may affect pain interpretation. The current study has several limitations. Professionals were asked to rate the pain experience of infants receiving heel sticks from videotape. Although this may approximate the real situation in a NICU setting, it is not identical. In a NICU setting, professionals would have rich information from the environment, previous contact with an infant, physiological data, and medical records that guide their assessment of pain. They would also view this infant within the context of all other infants in the unit. Professionals here were also asked only to provide ratings of pain intensity. As the results suggest, this is only one dimension of pain and may not be the dimension that plays the largest role in their judgments regarding pain in a clinical setting. The professionals here were experienced in the types of pain experienced in the NICU and may have held a priori beliefs about the painfulness of this procedure that moderated their judgments. Research suggests professionals' beliefs regarding the painfulness of a procedure play a large role in their assessments of children's pain [ 7 , 9 , 25 ]. Conclusions The current study indicates professionals' perception of the pain intensity of infants does not differ due to their understanding of the infants' level of risk for neurological impairment. Professionals also view cuddling as less effective for infants at greater risk for neurological impairment. Further research is needed to examine the reasoning behind the judgments made by healthcare professionals and to clarify why they might view an intervention as less effective for infants with greater risk of neurological impairment, despite having rated their pain intensity as similar to that of infants at lesser risk. Competing interests The authors declare that they have no competing interests. Authors' contributions The study was conceived and designed by BS, PM & LB with assistance from all remaining authors. The study was conducted under the supervision of PM, BS, KO, AO. Statistical analyses were conducted by LB, with assistance from PM, BS and JB. Interpretation of results were conducted by LB, PM, BS, JB, CC, LF, KB and AO. The manuscript was prepared by LB, and edited by PM and BS, with review and assistance from all remaining authors. All authors read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC534106.xml
524504	Comparison of Misoprostol and Dinoprostone for elective induction of labour in nulliparous women at full term: A randomized prospective study	Background The objective of this randomized prospective study was to compare the efficacy of 50 mcg vaginal misoprostol and 3 mg dinoprostone, administered every nine hours for a maximum of three doses, for elective induction of labor in a specific cohort of nulliparous women with an unfavorable cervix and more than 40 weeks of gestation. Material and Methods One hundred and sixty-three pregnant women with more than 285 days of gestation were recruited and analyzed. The main outcome measures were time from induction to delivery and incidence of vaginal delivery within 12 and 24 hours. Admission rate to the neonatal intensive care unit within 24 hours post delivery was a secondary outcome. Results The induction-delivery interval was significantly lower in the misoprostol group than in the dinoprostone group (11.9 h vs. 15.5 h, p < 0.001). With misoprostol, more women delivered within 12 hours (57.5% vs. 32.5%, p < 0.01) and 24 hours (98.7% vs. 91.4%, p < 0.05), spontaneous rupture of the membranes occurred more frequently (38.8% vs. 20.5%, p < 0.05), there was less need for oxytocin augmentation (65.8% vs. 81.5%, p < 0.05) and fewer additional doses were required (7.5% vs. 22%, p < 0.05). Although not statistically significant, a lower Caesarean section (CS) rate was observed with misoprostol (7.5% vs. 13.3%, p > 0.05) but with the disadvantage of higher abnormal fetal heart rate (FHR) tracings (22.5% vs. 12%, p > 0.05). From the misoprostol group more neonates were admitted to the intensive neonatal unit, than from the dinoprostone group (13.5% vs. 4.8%, p > 0.05). One woman had an unexplained stillbirth following the administration of one dose of dinoprostone. Conclusions Vaginal misoprostol, compared with dinoprostone in the regimens used, is more effective in elective inductions of labor beyond 40 weeks of gestation. Nevertheless, this is at the expense of more abnormal FHR tracings and more admissions to the neonatal unit, indicating that the faster approach is not necessarily the better approach to childbirth.	Introduction Induction of labor is carried out for maternal and fetal indications. One of the most common indications is prolonged pregnancy [ 1 ]. Recent studies have suggested that by continuing pregnancy beyond 41 weeks, there is a statistically significant higher perinatal morbidity and mortality as well as an increased risk to the mother [ 2 , 3 ]. Thus, there is a growing body of evidence suggesting the elective induction of labor at 41 weeks of gestation instead of expectant management [ 4 - 6 ]. Prostaglandin analogues, dinoprostone (PGE 2 ) and misoprostol (PGE 1 ), are widely used in "induction of labor" practice for ripening the cervix and stimulating uterine contractions in order to achieve vaginal delivery. Although dinoprostone has been approved by the FDA for cervical ripening in women at or near term, misoprostol is not currently approved for such use by the FDA, although it has the advantages of lower cost, no need for refrigeration and probably higher efficacy. Several studies have demonstrated a higher efficacy of vaginally administered misoprostol compared to vaginal dinoprostone for both cervical ripening and labor induction [ 7 - 19 ]. The Cochrane Pregnancy and Childbirth Group, having reviewed 45 randomized studies, concluded that vaginal misoprostol (25 to 100 mcg) was more effective than oxytocin or dinoprostone at the usual recommended doses for induction, but with increased rates of uterine hyperstimulation both with and without associated fetal heart rate (FHR) changes, as well as meconium stained fluid [ 9 ]. Most of the studies included in the previous meta-analysis used the 50 mcg dose for misoprostol at a maximum interval of six hours between the repeated doses, always resulting in higher rates of hyperstimulation. However, it is difficult to interpret previously published studies comparing misoprostol with dinoprostone for induction of labor since the majority of them are not double-blinded [ 9 ] and they have included both complicated and uncomplicated pregnancies, multiparous women with nulliparous as well as a wide gestational age (GA) range (37–42 weeks). Moreover, to reduce the risk of side effects, one can either decrease the dose of the drug [ 7 , 19 ] or prolong the dosage interval [ 14 , 15 , 19 ]. In addition, Alexander et al. have recently shown that in prolonged pregnancies it was not the induction per se that would increase the risk for caesarean section (CS), but patients related risk factors such as nulliparity and undilated cervix and the use of epidural analgesia [ 20 ]. This study was undertaken to compare the efficacy of vaginal misoprostol (50 mcg) with that of vaginal dinoprostone (3 mg) when both are administered at an interval of nine hours between repeated doses in a well-homogenized cohort of full term pregnancies (nulliparous women with an unfavorable cervix and without pregnancy complications). Material and Methods Between March 1, 2001 and July 16, 2003, 163 women were recruited for the study: 80 women in the misoprostol group and 83 women in the dinoprostone group. All of the women were recruited at Ioannina University Hospital, a tertiary referral center for high-risk pregnancies, with about 1600 deliveries a year. The Ethics Committee of the University of Ioannina approved the study and all participants gave their written informed consent after they had been made aware of the purpose of the study. Although the main indication was prolonged pregnancy, some of the inductions were performed at the patient's request after consultation at 40 weeks of gestation, (without any medical indications) and only if they had not delivered by the 285 th day of gestation. A sequence from a computerized random number generator was used for the allocation of patients to each group. The vaginal administration of prostaglandins was performed by one of the resident doctors on duty, who was not involved in managing these women in labor or delivery. The study was double blind, since the patients were not aware of which type of medication was used, and the deliveries were then performed by two gynecologists blinded to the induction regimen utilized. Inclusion criteria were: 1) age>18 years old 2) nulliparity, 3) accurate dating of gestation, including crown rump length (CRL) measurements in the first trimester of pregnancy, 4) singleton viable pregnancy, 5) gestational age ≥ 285 days, 6) cephalic presentation, 7) unfavorable cervical status defined as a Bishop score (BS) of ≤5, 8) intact membranes, 9) reactive non-stress test (NST). Exclusion criteria were: 1) known contraindications to receiving prostaglandins, 2) placenta previa, 3) prior uterine surgery and 4) any antenatal complications. GA was estimated by ultrasound biometry (via CRL measurements in the first trimester of pregnancy) in cases where there were more than 3 days difference from that obtained from the last menstrual period (LMP) [ 21 ]. Uterine tachysystole was defined as more than five contractions per 10 minutes, uterine hypertonus as when one contraction lasted more than 2 minutes and hyperstimulation syndrome as the presence of non-reassuring FHR tracing combined with either tachysystole or hypertonus. Non-reassuring FHR patterns were defined as persistent or recurring episodes of severe variable decelerations, late decelerations, prolonged fetal bradycardia or a combination of decreased beat-to-beat variability and a decelerative pattern [ 22 ]. A NST to ensure the well-being of the fetus was performed for each patient at the time of recruitment and admission to the hospital (at least 285 days of gestation) and one hour before the application of the prostaglandin. After the reassessment of the cervical BS, either 50 mcg misoprostol, or 3 mg dinoprostone was administered in the posterior vaginal fornix at 23:00 hours. The NST was repeated (duration of two hours) after 1 h and 5 h. If the woman was in active labor, the membranes spontaneously ruptured or the FHR not reassuring, the patient was transferred to the labor room. Otherwise, a second BS evaluation was carried out the next morning at 08:00 am (after 9 hours). If the cervix was favorable, (BS ≥ 5), the patient was admitted to the labor ward where oxytocin augmentation was carried out if the uterine contractions were unsatisfactory and amniotomy was performed when appropriate. If the cervix was still unfavorable, a second dose of misoprostol or dinoprostone was given and the same evaluation steps as described above were followed. After a total of 18 h had elapsed, non-responders were given a third dose of prostaglandin. When the third dose was insufficient for initiating spontaneous labor, a trial of labor was offered with oxytocin infusion and if no progress was achieved within 6 hours (based on digital assessment of the BS), the patient underwent a CS. The outcome measures were divided into "obstetrical" and "neonatal". The primary outcome measures were time from induction to delivery and incidence of vaginal delivery within 12 and 24 hours; the secondary outcomes were the CS rate, the need for oxytocin augmentation, the incidence of meconium stained amniotic fluid, the incidence of uterine tachysystole, abnormal FHR tracings, maternal morbidity, the admission to neonatal intensive care within 24 hours and neonatal arterial cord ph, base deficit. Statistical analysis was performed using SPSS version 11 software. The Chi square test and Fisher's exact test were used to analyze nominal variables in the form of frequency tables. Normally distributed (Kolmogorov-Smirnov Test with Lilliefors correction) metric variables were tested by the T-test for independent samples, while non-normally distributed metric variables were analyzed by the Mann-Whitney U test. All tests were two-tailed with a confidence level of 95% (p < 0.05). Values are expressed as mean ± standard error (SEM). Results The two groups were comparable in terms of patients' age (28.1 years vs.27.5, p > 0.05) and indication for induction (prolonged pregnancy 81.2% vs.78.3%, p > 0.05; social 18.8% vs. 21.7 %,) in the misoprostol and dinoprostone groups, respectively. Gestational age (286 days, range:285–292) and the preinduction BS (2.7 ± 0.1) in the misoprostol group were also comparable to the dinoprostone group (286 days, range:285–293) and (2.9 ± 0.1), respectively. Obstetrical outcome The induction-delivery interval was significantly shorter (11.9 h vs. 15.6 h, p < 0.001) in the misoprostol group, with even less need for a second or third dose (7.5% vs. 22%, p < 0.05) compared to dinoprostone. With misoprostol, more women delivered within 12 h (57.5% vs. 32.5%, p < 0.01) and almost all of the women delivered within 24 h (98.8% vs. 91.6%, p < 0.05). In addition, spontaneous rupture of the membranes occurred more often after the administration of misoprostol (p < 0.05) and there was a reduced need for oxytocin augmentation in labor: 65.8% vs. 80.7% with dinoprostone (p < 0.05). However, uterine tachysystole (p < 0.05)) and meconium stained amniotic fluid (p > 0.05) occurred more often in the misoprostol group as did abnormal heart rate tracing (22.5% vs.12%, p > 0.05) (Table 1 ). Table 1 Obstetrical Outcomes Misoprostol n = 80 (%) Dinoprostone n = 83 (%) Statistical significance Time from induction to delivery (h ± SEM) 11.9 ± 0.6 15.6 ± 0.7 p < 0.001 Delivery < 12 h 46 (57.5%) 27 (32.5%) p < 0.01 Delivery < 24 h 79 (98.8%) 76 (91.6%) p < 0.05 Number of doses Single dose 74 (92.4%) 65 (78.3%) p < 0.05 Second dose 6 (7.5%) 17 (20.5%) Third dose 0 (0%) 1 (1.2%) Required oxytocin augmentation 53(65.8%) 67(80.7%) p < 0.05 Spontaneous rupture of membranes 31 (38.8%) 17 (20.5%) p < 0.05 Meconium stained AF 15 (18.8%) 7 (9.6%) NS Abnormal FHR 18 (22.5%) 10 (12%) NS Uterine Tachysystole 10 (12.6%) 3 (3.6%) p < 0.05 Uterine Hyperstimulation 2 (2.5%) 1 (1.2%) NS FHR = fetal heart rate. AF = amniotic fluid NS = not significant (p > 0.05) SEM = standard error of the mean In both groups, the majority of women had vaginal delivery, 92.5% with misoprostol, and 86.7% with dinoprostone. There was no statistically significant difference between the two groups with regard to the CS rate (Table 2 ). There were no uterine ruptures or other major maternal complications resulting from the use of either of the prostaglandins. There was only one wound infection with dinoprostone, one woman in each group had delayed discharge due to persistent pyrexia and two women in the dinoprostone group required uterine packing (insertion of tampons within the uterine cavity) due to postpartum bleeding. Table 2 Mode of delivery and indications for Caesarean section Misoprostol n = 80 (%) Dinoprostone n = 83 (%) Statistical significance Vaginal 74 (92.5%) 72 (86.7%) NS 1 Spontaneous vaginal 46 (57.5%) 52 (62.6%) NS Vacuum assisted vaginal 28 (35.0%) 20 (24.1%) NS Caesarean section 6 (7.5%) 11 (13.3%) NS Nonreassuring FHR 2 4 (5.0%) 6 (7.2%) NS Failed induction 0 (0.0%) 1 (1.2%) NS Lack of labor progress 1 (1.3%) 2 (2.4%) NS Cephalopelvic disproportion 1 (1.3%) 2 (2.4%) NS 1 NS = not significant 2 FHR = fetal heart rate. Neonatal outcome More neonates in the misoprostol group had first minute Apgar scores lower than 7 (12.6% vs. 6.1%, p > 0.05), or needed neonatal resuscitation (11.4% vs. 9.9%, p > 0.05) but none of the babies had birth asphyxia [ 23 ]. The mean cord pH and the base deficit were comparable in the two groups. No neonate had meconium aspiration syndrome. Two neonates in the dinoprostone group had clavicle fracture (Table 3 ). There was no statistically significant difference in the number of neonates admitted to neonatal intensive care within 24 hours after delivery, between the misoprostol and dinoprostone groups (6.3% vs. 3.6% p > 0.05) (Table 4 ). Table 3 Neonatal Outcomes Misoprostol n = 80 (%) Dinoprostone n = 83 (%) Statistical significance Birth weight (g) 1 3275 ± 430 3373 ± 390 NS Perinatal death 0 1(1.2%) NS Neonatal resuscitation 9 (11.3%) 9 (10.8%) NS O 2 Supplementation 1 2 Ambou ventilation 7 6 Intubation in labor room 1 1 Apgar score < 7 1 min 10 (12.5%) 5 (6.0%) NS 5 min 1 (1.3%) 0 Cord blood pH (arterial) 1 7.28 ± 0.05 7.27 ± 0.05 NS Base deficit 1 5.0 ± 2.3 5.7 ± 3.2 NS 7.01 < cord pH < 7.20 3 (3.8%) 4 (4.8%) NS 10 < base deficit < 16 1 (1.3%) 2 (2.5%) NS Hyperbilirubinemia 2 9 (11.3%) 5 (6.0%) NS Birth trauma 3 0 2 (2.5%) NS 1 Values expressed as mean ± SD 2 Excluding pathological causes of icterus 3 Both were clavicle fractures Table 4 Admission to Neonatal Intensive Care Unit N (%) DA 1 Delivery 2 Indication Diagnosis HDs 3 Within 24 hours Misoprostol† 5 (6.3%) 01 VVD Rule out infection Elevated CRP 4 WBC 5 07 01 VVD Respiratory distress Respiratory infection 11 01 CS Rule out asphyxia Infection 10 01 SVD Respiratory distress Work up for infection 03 01 VVD Respiratory distress Work up for infection 04 Dinoprostone† 3 (3.6%) 01 VVD Respiratory distress Atelectasis 06 01 SVD Rule out asphyxia Infection 10 01 SVD Respiratory distress Respiratory infection 10 After 24 hours Misoprostol† 6 (7.2%) 02 VVD Rule out infection WBC in CSF 6 10 07 VVD Hyperbilirubinemia Urinary infection 07 04 SVD Hyperbilirubinemia Icterus 04 08 SVD Infection Respiratory infection 10 03 SVD Feeding difficulty WBC in CSF 20 05 CS Hyperbilirubinemia Icterus 04 Dinoprostone† 1 (1.2%) 17 SVD Fever WBC in CSF 15 1 DA = day of admittance 2 SVD = Spontaneous Vaginal Delivery, VVD = vacuum assisted vaginal delivery, CS = Caesarean section 3 HDs = hospitalization days 4 CRP = C-reactive protein 5 WBC = white blood cell 6 CSF = cerebrospinal fluid † not significant (p > 0.05) A 28-year-old woman at 41 weeks of gestation had an unexplained stillbirth after receiving a single dose of dinoprostone. Seven hours later she had a cardiotocogram without abnormal FHR patterns and regular contractions of the uterus were evident. We decided to move her to the labor ward and within half an hour of entry, no cardiac activity of the fetus was found. During these 30 minutes FHR monitoring had been discontinued, as it was not included in the study design. We decided to let her attempt vaginal delivery. An amniotomy was performed and the amniotic fluid was found to be clear and a vaginal delivery was achieved within 6 hours. Direct examination of the fetus, the placenta and the umbilical cord (UC) showed only a thin UC with excess twisting around its axis. The anatomopathology examination of the fetus revealed no abnormality except a microscopically decreased Wharton's jelly. Discussion Nowadays, induction of labor is more widely used than ever before [ 24 , 25 ]. Recent studies have shown that this increase is mainly due to a rise of inductions for marginal or elective reasons. The common indications are elective induction and postdate pregnancy often applied to gestations of 40 to 41 weeks [ 1 , 25 ]. Mongelli et al. have also shown that for the detection of post-maturity there is no advantage in using menstrual dates when ultrasound biometry is available [ 26 ]. Women may experience distress when labor has not started by the expected date [ 27 ] and obstetricians have to withstand pressure from these patients as well as the temptation to use prostaglandins earlier. Appropriate evaluation of the pregnancy and consultation with such patients will lead to the correct selection of those who will benefit most from a labor induction, thus eliminating the risk of post-maturity to the fetus without inducing fetal distress during labor. To the best of our knowledge, the present study is the only one that compares misoprostol and dinoprostone in such well-homogenized groups. All of the women were nulliparous with intact membranes and at more than forty weeks' gestation with no antenatal complications and all had an unfavorable cervix. In these carefully selected patients, misoprostol at the dose used not only shortened the time between induction and delivery (11.9 vs. 15.6 h), but it also was significantly more effective than dinoprostone. The positive point was that this result was achieved with a very low CS rate even in the dinoprostone group, (7.5%, and 13.3%), respectively. A difference of 5% in favor of misoprostol, although not statistically significant, might have clinical importance in terms of patient health and cost effectiveness. Although in the recent large meta-analysis [ 9 ] published by the Cochrane Library, the CS rates were inconsistent, they tended to be lower with misoprostol; an earlier study by Sanchez-Ramos et al. found a statistically significant difference in favor of misoprostol [ 28 ]. In addition, our results for this GA window are reassuring with regard to concerns that have been raised from previous retrospective studies reporting an increased risk of Caesarean delivery in nulliparous women when elective inductions are performed [ 29 , 30 ]. Even though misoprostol improves the kinetics of labor during induction in a more efficient way than dinoprostone, concerns persist with respect to intrapartum fetal "wellbeing". In order to avoid uterine hyperstimulation and abnormal FHR tracings, we used for first time in the literature, a 9 h interval between the prostaglandin doses. Although we indeed achieved a low rate of uterine hyperstimulation syndrome (2.5% with misoprostol and 1.2% with dinoprostone, respectively), we still noticed a trend towards a high rate of abnormal FHR tracings during induction with misoprostol. Our findings, in accordance with the previous Cochrane metanalysis [ 9 ], showed that with misoprostol there was an increased probability of meconium staining of amniotic fluid as well as of uterine tachysystole and of abnormal FHR tracings. In the misoprostol group, the majority of women also underwent either a CS or a vacuum operative delivery due to non-reassuring FHR. If neonatal outcomes such as neonatal resuscitation, low Apgar score in the first minute and admittance to the neonatal unit within the first 24 hours (none of the above were statistically significant but they were more frequent with misoprostol) are taken into account, misoprostol may increase these complications in labor. Thus, although our sample size cannot determine safety, misoprostol use is associated with a higher chance of admittance to the neonatal unit within 24 hours even in the absence of asphyxia. This evidence indicates that the faster approach to childbirth is not necessarily the better one. Attempting an explanation to the aforementioned side effects of misoprostol use and taking into account other reports [ 9 , 31 , 32 ], it appears that the increase in clinically relevant adverse effects is not only misoprostol related but it may be dose dependent. Lyons et al. have recently shown in term pregnant rats that a higher dose of misoprostol is needed to induce PGE2 secretion in the cervix than in the myometrium, and furthermore that EP3 receptors (prostaglandin E2 receptors) are differentially expressed in the myometrium (increased) than in the cervix (unaltered) in response to misoprostol [ 33 ]. The above findings indicate that misoprostol not only acts better on the myometrium than on the cervix, but an even higher dose is needed in order to ripen the cervix. Thus, it seems reasonable that increasing the interval between repeated misoprostol doses should reduce the risk of an asynchrony between a well or even hyper-stimulated uterus and a still not efficiently ripened cervix. Misoprostol probably has a large inter-patient variability in terms of pharmacokinetics, but it is also probable that the 50 mcg dosage may induce asynchrony between immature cervix effacement and uterine contractions, resulting in a more rapid but also more "stressful" labor. Based on these findings, we would propose, in future, a slight modification of the misoprostol protocol used in this study. An initial lower dose of misoprostol (20–25 mcg), followed by 50 mcg should be considered in trying to achieve priming of the cervix without inducing such high uterine contractility and neonatal complications. Indeed, in a recent study comparing 25 mcg misoprostol with 1 mg dinoprostone administered vaginally every four hours, the admission rate to neonatal intensive unit was significantly lower in the misoprostol group [ 34 ]. It still has to be mentioned that in many of our participants, the vertex was not engaged in the pelvic inlet on the day of admittance and this should have been included as an independent risk factor in the initial study design. The exact cause of the stillbirth in the dinoprostone group remains unclear, emphasizing thus, the need for continuous FHR monitoring during labor induction if regular uterine contractions persist [ 35 , 36 ]. Conclusions To conclude, 50 mcg misoprostol at a 9 h interval is more highly effective in promoting cervical ripening and in inducing labor, compared to dinoprostone. However, certain aspects concerning fetal well being during labor induction remain questionable. Larger prospective studies comparing elective induction to expectant management after a completed 40-week gestation (on the basis of early ultrasound biometry) might reveal a subgroup of women, such as nulliparous with an unfavorable cervix, who might benefit from an elective induction, preferably with a 25 mcg misoprostol initial dose. Authors' contributions E.P: conceived of the study, participated in the sequence alignment, performed the statistical analysis and drafted the manuscript. N.P: conceived of the study, and performed the labor inductions A.D: performed the neonatal examination and follow up S.A: performed the neonatal examination and participated in the neonatal data analysis C.V: was the midwife involved in women allocation, medications preparation and labour data registration T.S: conceived of the study and data analysis E.P: conceived of the study and coordinated the study K.Z: conceived of the study, performed the labor inductions and coordinated the study	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC524504.xml
520832	The C677T methylenetetrahydrofolate reductase variant and third trimester obstetrical complications in women with unexplained elevations of maternal serum alpha-fetoprotein	Introduction The C677T MTHFR variant has been associated with the same third trimester pregnancy complications as seen in women who have elevations of maternal serum α-fetoprotein (MSAFP). We hypothesized that these women with third trimester pregnancy complications and MSAFP elevations would have an increased frequency of the variant compared to an abnormal study control group (women with MSAFP elevations without pregnancy complications) as well as to normal population controls. Methods Women who had unexplained elevations of MSAFP in pregnancy were ascertained retrospectively. The frequency of the C677T MTHFR variant among those women with unexplained elevations of MSAFP who had experienced later pregnancy complications was compared to that of women with unexplained elevations of MSAFP without complications as well as to that of the previously established Manitoba frequency. Results Women who had complications of pregnancy and an unexplained MSAFP elevation had a higher allele frequency for the C677T MTHFR variant (q = 0.36,) compared to women with MSAFP elevations and normal pregnancy outcomes (q = 0.25, OR 1.73 95% CI 1.25–2.37, p = 0.03). The frequency was also higher than that of the population controls (q= 0.25, OR 1.70 95% CI 1.11–2.60, p = 0.007). The frequency in women with MSAFP elevations without pregnancy complications was not significantly different from that of the population controls ( p = 0.41). Conclusion Women with unexplained elevations of MSAFP and who experience complications in later pregnancy are more likely to have one or two alleles of the C677T MTHFR variant.	Background Significant elevations of amniotic fluid and maternal serum alpha-fetoprotein (MSAFP) have been shown to be associated with spina bifida and other neural tube defects (NTD). The province of Manitoba, Canada offers province-wide midtrimester MSAFP screening to all pregnant women. It has been recognized that some pregnant women with midtrimester unexplained elevations of MSAFP [ 1 , 2 ]. Increased total plasma homocysteine alters placental function and has been associated with the same complications that are associated with unexplained elevated MSAFP [ 3 - 7 ]. The C677T MTHFR variant has also been associated with complications of pregnancy in some, but not all, studies [ 8 - 10 ] C677T MTHFR may therefore contribute to complications of pregnancy by elevating serum homocysteine. Poor placental function could result in both an unexplained elevation of MSAFP and complications of pregnancy. We therefore hypothesized that women with third trimester pregnancy complications and MSAFP elevations (cases) would have an increased frequency of the variant compared to the Manitoba population (population controls) or women with MSAFP elevations without pregnancy complications (study controls) if they had low folate intake. Methods Background to methods In a small pilot study of 32 couples, we found that women who had an unexplained elevation of MSAFP and a normal midtrimester fetal ultrasound, and their partners, had a significantly increased C677T MTHFR frequency compared to Manitoba newborns (RR 1.42, 95% CI 1.08–1.85, p = 0.012, two tailed) [ 11 ]. The newborn study that examined 977 anonymous consecutive neonatal screening blood spots showed that 36% of Manitoba newborns were heterozygous and 7% were homozygous for C677T MTHFR [ 12 ] (q = 0.25). Subsequently, on evaluation of the pregnancy outcomes of our pilot study women, we noted that, among eight women who had gone on to experience complications of pregnancy, the odds ratio for having the C677T MTHFR allele was 2.3 times higher than in the Manitoba population. However, the result was not statistically significant ( p = 0.151, two tailed) indicating the frequency was increased but, this could have been a random result. Ascertainment and recruitment of study population All pregnant women in Manitoba are eligible for routine serum screening through the voluntary MMSSP. In Manitoba, an elevation of MSAFP is defined as 2.3 multiples of the median (MOM) or greater. Candidates for inclusion in this study were women with an unexplained MSAFP elevation (i.e. not due to fetal anomalies, incorrect estimation of gestational age, previously unrecognized fetal demise, or multiple gestation) with either a complicated or uncomplicated pregnancy outcome. After appropriate approvals had been obtained from The University of Manitoba Health Research Ethics Board, review of the screening records began in 1999 and took three years. For a study using a two step consent to participate methodology administered by mail, the expected response rate (after excluding lost to follow-up) would be 20% [ 13 ]. Our goal was 1000 invitations. We anticipated this would result in approximately 120 participants. This would be double the minimum number of participants suggested by the power analysis we had conducted for the pilot study. To increase our response rate further, we added telephone follow-up for invited potential participants who were non-responders [ 14 ]. All screening records from 1995–1999 were reviewed, accounting for 783 invitations. Records for 2000–2002 were reviewed systematically as outcome information on each pregnancy became available to MMSSP. Records for 1990–1994 were then reviewed systematically in order to bring the total up to 1000. If a woman had more than one pregnancy with an elevation of MSAFP screened by the MMSSP, only the first pregnancy encountered in the retrospective review was used for the study. Previous or subsequent pregnancies were not included. Women with preexisting conditions known to influence pregnancy outcome, such as essential hypertension, and mothers of babies with major congenital anomalies were excluded. Eight women who had relinquished their babies for adoption or whose babies were placed in foster care were also excluded. Women who met the inclusion criteria were divided into two groups for analysis. Cases were defined as women with pregnancies complicated by one of the complications previously shown to be associated with an unexplained elevation of MSAFP at midtrimester [ 2 ]. These include: intrauterine growth restriction (IUGR) (<10 th percentile), pregnancy induced hypertension, preeclampsia, eclampsia, postpartum hemorrhage, retained placenta requiring manual delivery, abruptio placenta, premature delivery (<36 weeks gestation or requiring specialized neonatal care for prematurity) and unexplained fetal demise. Study controls were women with normal outcomes which were defined as those with delivery at term ≥ 36 weeks gestation), no complications of pregnancy, a normal placenta and a healthy baby. Definition of complications was based on ICDC-9 codes in the MMSSP outcome charts for each patient [ 15 ] which are then confirmed later by chart review for all those with a positive MMSSP result. All women ascertained as having unexplained MSAFP elevations and who fit the inclusion criteria above, were invited by letter to participate. The previously reported newborn study provided population control group data [ 12 ]. Study questionnaires Women who agreed to participate in the study were mailed the appropriate questionnaires and blood requisitions. The questionnaire included a semi-quantitative food frequency questionnaire (FFQ) based on standard methodology but, modified to suit Manitoba residents and previously validated for this population by biochemical analysis during the pilot study [ 11 , 16 ]. The survey included questions on vitamin supplement intake to determine preconceptional or prenatal supplementation as well as current use of vitamins. Dietary intake of folate and folic acid from supplements, and intake of the cofactors B 12 and B 6 , were calculated from the FFQ for intake both during pregnancy and at the time of the study. A correction of an additional 0.1 mg for folic acid fortification that began in Canada in 1998 was included for pregnancies that began after fortification [ 17 ]. FFQ analyses were performed with the researcher blinded as to the status of the mother. Laboratory analysis Total plasma homocysteine, red blood cell folate, and serum folate were determined using established methodology [ 18 , 19 ]. Samples were processed on site with clotting and separation by spinning. Sera was stored at 4°C during shipping to the central laboratory and until processing. DNA was extracted from whole blood and C677T MTHFR genotyping was performed using previously established methodology [ 11 , 20 , 21 ]. Genotyping and biochemical analyses were performed also blinded. Statistical analysis Chi-squared analysis (one tailed unless otherwise noted) was used for allele frequency. Comparisons of potentially confounding factors between the case group and the study control group were undertaken. Parametric data were analyzed with the Student's t test for difference between means with Bonferroni correction for multiple comparisons. Data not normally distributed were analyzed using the nonparametric Mann-Whitney Rank Sum Test. Linear regression was used to test the validity of the dietary survey. A multivariate analysis included age, smoking, maternal weight at the time of MSAFP testing, presence of C677T MTHFR, gender and weight of infant, biochemical parameters, and FFQ results for folate, B 12 and B 6 , both at the time of the survey and for during the pregnancy was undertaken. In order to avoid convergence due to the large number of variables, the analysis was completed in subsets of six variables. Variables with the higher association scores from these analyses were then combined for further testing in various combinations using stepwise multiple linear regression. Also linear regression analysis of each continuous variable with genotype results was performed. Corrections for multiple comparisons were included. Software used was NCSS Statistical Systems for Windows [ 22 ]. Results Participation rates Nine hundred and ninety four women were identified as eligible (342 cases and 652 controls). Of the 590 women successfully contacted, 130 (22%) agreed to participate (56 cases and 74 controls). Four hundred and four women were lost to follow-up. Cases were more likely to choose to participate than controls and this difference was significant (1.5, p = 0.030). There was no difference in the proportions of cases and controls that were lost to follow-up ( p = 0.157). We had anticipated a 20% response rate and we achieved 24%. Genotype results Genotypes were available for 54 cases and 73 controls for this analysis. Results are summarized in Table 1 . The allele frequency for the C677T MTHFR variant in the Manitoba population has been previously established to be q = 0.25. Women who had complications of pregnancy and an unexplained MSAFP elevation had a higher allele frequency for the C677T MTHFR variant (q = 0.36) compared to women with MSAFP elevations and normal pregnancy outcomes (q = 0.25, OR 1.73 95% CI 1.25–2.37, p = 0.03). The frequency was also higher than in the population controls (q = 0.25, OR 1.70 95% CI 1.11–2.60, p = 0.007). The frequency in women without pregnancy complications and MSAFP elevations (study controls) was not significantly different than that seen in population controls ( p = 0.41). Table 1 Comparison of allele frequency of C677T MTHFR between cases, study controls, and population controls. Subjects C/C (%) C/T (%) T/T (%) Comparing to study controls* OR (95%CI) Comparing to population* OR (95%CI) Cases N = 54 21 (39) 27 (50) 6 (11) 1.73 (1.25–2.37) ( p = 0.033) 1.70 (1.11–2.60) ( p = 0.007) Study Controls N = 73 40 (55) 30 (41) 3 (4) ~ 0.98 (0.46–1.55) ( p = 0.410) Population N = 977 557 (57) 352 (36) 68 (7) 0.98 (0.46–1.55) ( p = 0.410) ~ * χ 2 comparison of allele frequency (total T and C) in each group, one tailed. Cases: women with unexplained elevations of MSAFP who had subsequent complications of pregnancy, (C = 69, T = 39) Controls: women who had unexplained elevations of MSAFP and no subsequent complications (C = 110, T = 36) Population controls were 977 newborns (C = 1466, T = 488) [12]. C/C = normal type, C/T = heterozygous for thermolabile variant, T/T = homozygous for thermolabile variant. The case and study control groups included women at various stages of their child bearing years. Only one woman recruited as a control subject had a previous or subsequent pregnancy with an unexplained elevation of MSAFP and complications. She was a heterozygote for C677T MTHFR. No case subjects had a previous or subsequent pregnancy with an unexplained elevation of MSAFP and a normal outcome, but four case subjects had had a previous or subsequent pregnancy with complications after an elevated MSAFP. If the case versus control classification had been based on whether or not a woman had ever had a pregnancy with an unexplained elevation of MSAFP followed by complications, the association would still be present when compared to the population control (q = 0.3636, OR 1.72, 95%CI 1.27–2.61, p = 0.0055). Biochemical results Heterozygotes and homozygotes for C677T MTHFR had lower average values (r = 0.978, p = 0.019) for serum folate than those who did not have the variant. None of the women were deficient in either serum folate (defined as <7.0 nmol/L) or red blood cell folate (defined as <430 nmol/L RBC). There was no significant difference in mean homocysteine levels (Table 2 ). Table 2 Comparison of the parametric characteristics of women with unexplained elevations of MSAFP according to those with and without complications of pregnancy Characteristic Mean Cases (SD) Mean Controls (SD) p value MSAFP result 2.78 (± 0.62) 3.16 (± 3.76) 0.398 weeks gestation 17.1 (± 1.61) 16.9 (± 1.40) 0.432 μg/folate/day in pregnancy 2 1216 (± 915) 1010 (± 892) 0.206 μg/folate/day at time of study 2 557 (± 341) 523 (± 498) 0.588 erc folate (nmol/L RBC) 1234 (± 289) 1208 (± 317) 0.632 serum folate (nmol/L) 32.3 (± 5.80) 32.3 (± 5.71) 0.956 serum homocysteine (μmol/L) 7.8 (± 2.26) 8.4 (± 2.80) 0.246 μg B 12 /day in pregnancy 2 12.4 (± 5.37) 13.4 (8.31) 0.488 μg B 12 /day at time of study 2 8.9 (± 12.26) 8.6 (± 10.83) 0.899 mg B 6 /day in pregnancy 2 8.4 (± 9.69) 7.2 (± 9.01) 0.461 mg B 6 /day at time of study 2 6.0 (± 13.35) 5.5 (± 11.10) 0.792 mother's age at delivery 31 (± 4.19) 30 (± 5.19) 0.251 mother's weight in Kg 76 (± 17.26) 69 (± 16.09) 0.013 1 1 This value is not significant after Bonferroni correction for multiple comparisons. See discussion. 2 Data was skewed due to a small number of women in both groups taking large dose vitamin supplements. When these women were removed from the analysis the result remained nonsignificant. Validity of surveys Seven cases and one study control declined to fill out their dietary surveys. Mean values were inserted in the multivariate analysis for these eight women. The validity of the dietary survey was demonstrated again for this study by linear regression analysis. Consistent with known homocysteine metabolism [ 23 ], a negative correlation existed between serum homocysteine and both red blood cell folate (r = 0.945 p = 0.0052) and serum folate (r = 0.932, p = 0.0001). Higher intake of dietary folate (including synthetic folic acid from supplements) as reported by the FFQ for the time of study was associated with higher serum folate (r = 0.941, p = 0.0001) and higher red blood cell folate (r = 0.949, p = 0.0166). We did several checks to determine that the women were answering their surveys accurately. Comparisons of specific data items available in the women's MMSSP charts at the time of pregnancy with the data reported in the surveys showed excellent agreement for every item examined indicating women answered questions accurately. Women who reported smoking (as a quantitative value from 0–3 based on 1/2 packs/day smoked) showed a negative correlation with serum folate (r = 0.923, p = 0.0062) consistent with accurately reporting their smoking habits [ 24 ]. Based on the results of these tests of the validity of our surveys, we are confident that the information provided by our participants was accurate. Analysis of FFQ survey and MMSSP data The ethnicity of the infants born to the case mothers (based on the ethnicity reported for the infants grandparents) was 84% Caucasian, 5% Aboriginal. Mixed ethnicity was reported for 11% of the infants with one parent Caucasian and the other parent Aboriginal, or rarely Black or Asian. The ethnic distribution was the same for controls and is typical for the Manitoba population [ 25 , 26 ]. There were also no significant differences between cases and controls with respect to their place of residence within the province (such as rural versus urban address). There were no significant differences in dietary and supplemental intake of folate, B 12 , or B 6 , or in the biochemical parameters of case and control mothers. There was no difference in the percentage of cases and controls who reported taking prenatal vitamin supplements during pregnancy (37/48 cases and 55/72) or taking vitamin and/or folate supplements preconceptionally (17/48 cases and 25/72 study controls). We attempted to divide our cases into smaller groups by type of pregnancy complication. We also separated isolated IUGR and IUGR associated with hypertensive disorders of pregnancy. Most of the groups lacked power for statistical analysis due to small numbers. However, normotensive women whose fetus had IUGR (N = 12) had a higher frequency of the C677T MTHFR variant compared to the population controls (q = 0.33, OR 2.58 95% CI, 1.78–3.73, p = 0.013). Homozygosity for the C677T MTHFR variant is associated with IUGR in women who do not take vitamin supplements according to one large study of Canadian women [ 10 , 27 ]. Our findings are in agreement with this result as only 3/12 women took supplements. We found this effect in a group of combined heterozygous and homozygous women. There was a trend towards higher mean weight for mothers who had complications at the time of MSAFP test in the individual comparisons, but this was not significant after correction for multiple comparisons (Table 2 and 3 ). The multivariate analysis did not reveal any unexpected associations, but it did show the importance of maternal weight as a variable (r = 0.933, p = 0.024). This was also not unexpected given that some of the complications we were examining are associated with obesity [ 28 ]. Even after controlling for women's weight in the multivariate analysis, the higher frequency of C677T MTHFR among cases remained significant (r = 0.734, p = 0.0462). There was no association between weight and MTHFR status (r = 0.431, p = 0.679). Table 3 Comparison of the nonparametric characteristics of women with unexplained elevations of MSAFP according to those with and without complications of pregnancy. Nonparametric Characteristics Cases Controls p value Location Winnipeg 44 63 0.854 southern city 3 2 southern town 3 5 southern rural 8 8 northern city 2 1 northern town 1 4 northern rural 5 7 ethnicity 0.972 Caucasian 52 66 Aborginal 2 2 Mixed Caucasian/Aboriginal 7 6 Mixed Caucasian/Black 1 3 Asian 1 1 Mixed Caucasian/Asian 1 1 Unknown 3 1 diabetes in pregnancy 2/67 1/80 0.207 maternal smoking present (0 = nonsmoker, 1–3 = half pks/day increments) 0 = 47, 1 = 11, 2 or more = 9 0 = 60, 1 = 8, 2 or more = 12 0.721 gender of baby 27 females, 40 males 42 females, 38 males 0.352 parity = number of women 0 = 37, 1 = 20, 2 = 7, 3 or more = 3 0 = 40, 1 = 29, 2 = 7, 3 or more = 4 0.254 previous miscarriages 11/67 15/80 0.203 previous case pregnancy 1 0 0.967 Discussion Unexplained elevations in MSAFP are known to be associated with an increased risk for complications of pregnancy [ 2 ]. Others have reported that presence of the C677T MTHFR variant in pregnant women with low folate intake is associated with increased risk for pregnancy complications [ 2 , 29 - 31 ]. The unique finding of this study is an increase in the frequency of the C677T MTHFR variant among women with normal folate intake, who went on to have complications of pregnancy after an unexplained elevation of MSAFP (Table 1 ). The lack of folate deficiency in this population was unexpected, given previous research which showed that 23.6% of Newfoundland and Labrador women are folate deficient at their first prenatal visit [ 32 ]. As our study was retrospective, we did not have data on levels during pregnancy. It has recently been shown that the C677T MTHFR variant does not affect maternal serum homocysteine levels in pregnancy among women who take prenatal multivitamins [ 8 ]. Also a recent prospective study shows that there is no difference in homocysteine levels at midtrimester between women who later develop preeclampsia and those who do not [ 33 ]. As is the situation with NTDs, lack of folate deficiency by current definitions in a non-pregnant woman may not indicate that her folate intake is adequate for pregnancy. This would especially be true for women with the C677T MTHFR variant. Reexamination of the current definition of what constitutes a normal biochemical result for folate intake for women of child bearing age should be undertaken to clarify this. We suggest that the negative effects of the C677T MTHFR variant are more likely to occur in early pregnancy before women began taking prenatal vitamins because the majority of our study participants took prenatal vitamins, but only 35% took preconceptional supplements. We suspect that reduced methylation interfering with cell proliferation in the placenta as originally suggested by Eskes (2000) [ 3 ]. In conclusion, using a retrospective case/control study, we have found that women with unexplained MSAFP elevations who have complications in later pregnancy are more likely to have the C677T MTHFR allele. Our resultsdo not suggest that C677T MTHFR predisposes a woman to having an elevation of MSAFP level (as we did not compare the C677T MTHFR frequency in women with and without elevated MSAFP), but having one or more copies of this variant predisposes such screen positive women to having complications in later gestation. It remains to be seen if other risk factors can be identified which can more accurately define this high risk group. Authors' contributions All authors participated in original study design except CS. CG, BC and CS all acted as principal investigators for funding. NB assisted with all grant proposals. NB undertook the review of the individual MSAFP files and MMSSP database searches, designed the dietary and family history surveys, classified cases and controls, acted as study coordinator handling all aspects of participant contact, and provided data analysis. All authors also participated actively with NB for various aspects of the study. CG provided MTHFR genotyping. LS provided biochemical analysis. NB drafted the original manuscript with assistance from BC. CG and BC handled ethics approval assisted by NB. All authors read and approved the final manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC520832.xml
517720	Vitamin D deficiency and causative factors in the population of Tehran	Background There are multiple studies in different countries regarding the prevalence of vitamin D deficiency. These studies showed high prevalence of vitamin D deficiency in Asian countries. This study tries to elucidate the prevalence of vitamin D deficiency and its influencing factors in population of Tehran. Methods 1210 subjects 20–64 years old were randomly selected. 25 (OH) D serum levels were measured. Duration of exposure to sunlight, the type of clothing and level of calcium intake and BMI were quantified based on a questionnaire. Results A high percentage of vitamin D deficiency was defined in the study population. Prevalence of severe, moderate and mild Vitamin D deficiency was 9.5%, 57.6% and 14.2% respectively. Vitamin D serum levels had no significant statistical relation with the duration of exposure to sunlight, kind of clothing and BMI. Calcium intake in the normal vitamin D group was significantly higher than the other groups (714.67 ± 330.8 mg/day vs 503.39 ± 303.1, 577.93 ± 304.9,595.84 ± 313.6). Vitamin D serum levels in young and middle aged females were significantly lower than the older group. Conclusions Vitamin D deficiency has a high prevalence in Tehran. In order to avoid complications of vitamin D deficiency, supplemental dietary intake seems essential.	Background Vitamin D is an essential element for establishing and maintananing bone structure. Vitamin D deficiency results in rickets and osteomalacia. Even slight vitamin D deficiency results in secondary hyperparathyroidism and increased bone resorption [ 1 , 2 ]. In addition, there has been increased attention to the physiologic importance of vitamin D in non-skeletal tissues [ 3 ]. Vitamin D is supplied by consumption of vitamin D-rich foods and by vitamin D synthesis in skin. Natural nutrient materials are not a sufficient source of vitamin D to supply the body requirements; therefore where there is no supplementation of foodstuffs, the main source for vitamin D is produced by UV light [ 4 , 5 ]. Regarding the significant role of sunlight in vitamin D synthesis, it is quite logical to suggest low prevalence of vitamin D deficiency in tropical countries. However the studies carried out in the preceding two decades have shown a high prevalence of vitamin D deficiency in tropical countries such as China, Turkey, India, Iran and Saudi Arabia [ 6 - 14 ]. The prevalence of vitamin D deficiency varied between 30% and 93%. However, the majority of these studies were limited to specific age and sex groups. Therefore, elucidation of vitamin D status at the community level and in different climates of a country seems essential. The present study is a part of a national project of prevention, diagnosis and treatment of osteoporosis that investigates the prevalence of vitamin D deficiency and its influencing factors in the population of Tehran. Methods 1272 healthy men and women aged 20–69 years were selected based on randomized clustered sampling from 50 blocks in Tehran. Exclusion criteria were known hepatic or renal disease, metabolic bone disease, malabsorption, sterility, oligomenorrhea, type I diabetes, hypercortisolism, malignancy, immobility for more than one-week, pregnancy, lactation, and medications influencing bone metabolism. The study protocol was approved by research ethics committee of Endocrinology & Metabolism Research Center (EMRC). Sampling was performed after taking informed consent at the beginning of 2001 in the subjects place of residence. 1210 of 1272 selected subjects participated in this study (response rate was 95%). One fasting blood sample was taken from each participant in his/her place of residence. Sample centrifuge and serum extraction were done in the field. Then samples were sent to the EMRC laboratory for analysis and were frozen immediately. 25-hydroxy vitamin D (25(OH) D) level was measured with RIA method (Biosource Europes.A,Ò). Normal range for serum vitamin D (25(OH) D) was 23 to 113 nmol/l. Serum PTH measurement was done using RIA method (Diasorin,Ò). Normal range for PTH is 13 to 54 nmol/l. Interassay and Intrassy for 25(OH) D were 8%, 6.8% and for PTH were 8.9% and 6.1% respectively. The subjects were asked to complete a questionnaire at the time of bone mineral densitometry analysis. The questionnaire included details of duration of exposure to sun light in previous month (less than 30 minutes/day; between 30 to 60 minutes/day; between 60 to 120 minutes/day; more than 120 minutes/day), sunscreen cream usage, clothing (exposure of hand and face or more than). In order to quantify the level of vitamin D and calcium consumption in the previous month, a food frequency questionnaire which was designed and standard by the Iranian Nutrition Institute was completed. Height and weight were measured at this stage. 25(OH)D equal or less than 12.5 nmol/l was considered as severe vitamin D deficiency or group 1 and vitamin D more than 12.5 nmol/l and less than 25 nmol/l was considered as moderate deficiency or group 2 [ 15 ]. PTH changes in various vitamin D serum levels were applied to detect mild vitamin D deficiency which has 25 (OH)D more than 25 nmol/l and less than or equal to 35 nmol/l. Threshold for mild vitamin D deficiency was measured by applying PTH changes in different serum levels of 25(OH) D. SPSS software (version 10) was used for data analysis. In descriptional statistics 5, 50, 95 percentiles were used. Results were expressed as mean ± SD or median. To find any significant difference between groups, X 2 test Kruskal-Wallis were used. Results In order to quantify serum levels of vitamin D and other biochemical parameters, serum samples were taken from 1210 subjects (response rate was 95%). 41% of subjects were male and 59% were female. Age and sex distribution of participants are shown in table 1 . In the second part of study (recall for bone mineral densitometry) for 666 subjects the questionnaire was completed. Table 1 Age and Sex distribution of participants Age(year) Total number Female Male 20–29 241 128 113 30–39 308 203 105 40–49 294 191 103 50–59 209 116 93 60 > 158 77 81 Figure 1 demonstrates vitamin D levels histogram in the study population. Total prevalence of severe, moderate and mild vitamin D deficiency was 9.5 %, 57.6% and 14.2 % respectively (Figure 2 ). Figure 1 Histogram of Vitamin D serum levels in study population Figure 3 demonstrates 95, 50 and 5 percentiles of vitamin D according to age and sex. Figure 3 Median, 5, and 95 percentile of Vitamin D in variable age and sex groups Serum levels of vitamin D in females above 60 years was higher than in other age groups (P < 0.001: Kruskal-Wallis test). Vitamin D serum levels in females between 20–29 years and 30–39 years was less than other age groups (P < 0.001). Median vitamin D level in females in age range of 20–29 years and above 60 years was 17 nmol/l and 39 nmol/l, respectively. Prevalence of high level of vitamin D (more than 150 nmol/l) in 60–69 years old female age group was significantly more than other age and sex group (P < 0.01). In recalling for bone densitometry, 666 returned (55% of study population) in whom the effect of influencing factors was evaluated. Table 2 shows mean BMI and daily calcium intake in different vitamin D groups. BMI was not significantly different in vitamin D groups but calcium intake in normal vitamin D group was significantly higher than other groups. Table 2 Mean BMI and daily calcium intake in variable vitamin D groups Groups Parameters Vitamin D ≤ 12.5 (nmol/l) 12.5<vitamin D ≤ 25 25<vitamin D ≤ 35 35.1<vitamin D ≤ 150 BMI (kg/m 2 ) 27.32 ± 5.02 26.44 ± 4.52 27.66 ± 5.16 26.99 ± 4.93 Calcium intake (mg/day) 503.39 ± 303.1* 577.93 ± 304.9* 595.84 ± 313.6* 714.67 ± 330.8 *Significant difference with normal group (35.1<vitamin D ≤ 150) P < 0.05 Discussion In our study the prevalence of severe and moderate vitamin D deficiency was 9.5 % and 57.6%, respectively. Mild vitamin D deficiency had a prevalence of 14.2%. Multiple studies have been carried out about the prevalence of vitamin D deficiency but they were mostly limited to a small sample size or assessed a specific age group (especially elderly). In countries where vitamin D fortified foodstuffs are available (USA and some Scandinavian countries), prevalence of vitamin D deficiency is between 1.6–14.8% in different age groups [ 16 - 18 ]. In other European countries where there is no vitamin D supplementation, deficiency is more prevalent. The studies which assessed middle-aged and elderly people showed vitamin D deficiency prevalence of 14% to 59.6% in these age groups [ 19 - 22 ]. Vitamin D deficiency prevalence is much higher in Asian countries. Fonseca and colleagues, demonstrated vitamin D level above 10 ng/ml in only 3 saudian females out of 31 [ 13 ]. Sedrani and colleagues showed vitamin D deficiency prevalence of 44%–100% in Saudian young females with different coverage and race [ 9 , 10 ]. Azizi & colleagues showed vitamin D level less than 18 ng/ml in half of the study population. Vitamin D deficiency prevalence in 10–19, 20–24, 30–41 was 47.4%, 59.5%, 44.8% respectively [ 11 ]. In the present study 81.3 % of subjects had vitamin D deficiency. Most studies have shown higher prevalence of vitamin D deficiency in the elderly [ 15 - 18 ]. Elderly females demonstrated statistically significant higher serum levels of vitamin D compared with young and middle aged females. Parenteral vitamin D intake by elderly was the major differentiating factor between various age groups that could explain high prevalence of a high level of vitamin D in elderly females. Subjects who took vitamin D in the sampling period were excluded from the study, but those who had taken vitamin D in the preceding months were not omitted. Vitamin D has a long half-life and its frequent prescription especially in elderly women with musculo-skeletal complaints can explain differences in serum vitamin D. Regarding the essential role of sunlight in vitamin D synthesis, it is quite unexpected to see a high prevalence of vitamin D deficiency in countries such as Saudi Arabia. Different hypotheses can be made such as insufficient sun exposure, clothing habits, hyper pigmentation, air pollution, insufficient intake of vitamin D and special dietary habits [ 27 ]. Although sunlight plays an essential role in vitamin D synthesis, its' role in vitamin D deficiency of Asians is not obvious. Tehran, which is located in 36° 21''N, has a mean sun exposure of 8 hours per day [ 28 ]. In the present study sun exposure was not significantly different between subjects with vitamin D deficiency and those with normal vitamin D status. Although there is sufficient sunlight in all seasons in Saudi Arabia, Sedrani showed that half of people who had more than 30 minutes of sun exposure had vitamin D less than 8 ng/ml (20 nmol/l) [ 10 ]. Holick & colleagues showed similar rate of vitamin D synthesis in Asians as of Europeans; but Asians required greater duration of exposure [ 29 ]. Other studies showed the same degree of increase in 25 (OH) D in summer months in Asians compared with Europeans [ 30 ]. In our study, there was no difference in clothing habits of vitamin D deficient group and normal group. Sedrani showed 70% vitamin D deficiency in males compared with30 % in young females in spite of greater clothing in females [ 10 ]. Another hypothesis says that air pollution prevents enough UV exposure to skin. Insufficient vitamin D intake is another hypothesis for high prevalence of vitamin D deficiency in Asians. Insufficient dietary supplies of vitamin D in countries where foodstuffs are not supplemented, leads to generally low dietary intake of vitamin D. In the Omdahl study, daily vitamin D intake in elderly healthy women was 54 units [ 16 ]. Our study does not assess daily dietary vitamin D intake. Decreased dietary calcium level induces increased serum PTH level and increased catabolism of 25 (OH) D, therefore decreased 25(OH)D is induced by dietary calcium deficiency [ 15 ]. Average calcium intake was 660 ± 350 mg/day in this study. There was no significant difference in dietary calcium intake among the different vitamin D groups. Although consumption of phytates and animal-derived proteins was not investigated in present study, high dietary consumption of phytates and low dietary intake of animal proteins is one of the suggested hypothesis for vitamin D deficiency [ 24 , 26 , 27 ]. There are other hypotheses to explain vitamin D deficiency among Asians. Awumey et al showed higher activity level of 24-hydroxylase in fibroblasts of Indian-Americans compared with controls [ 31 ]. Therefore, increased vitamin D catabolism may cause vitamin D deficiency in Asians. In order to elucidate specific etiologies responsible for high prevalence of vitamin D deficiency in Asians further studies should be carried out. It is possible that vitamin D deficiency is induced by combination of above mentioned etiologies. In order to clarify the significance of each etiologic factor, randomized controlled trials are necessary. Conclusions Given the high prevalence of vitamin D deficiency in Iran, effective solution to overcome its consequences seems indispensable. Competing interest None declared. Figure 2 Frequency of variable Vitamin D groups Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC517720.xml
533866	Itch and skin rash from chocolate during fluoxetine and sertraline treatment: Case report	Background The skin contains a system for producing serotonin as well as serotonin receptors. Serotonin can also cause pruritus when injected into the skin. SSRI-drugs increase serotonin concentrations and are known to have pruritus and other dermal side effects. Case presentation A 46-year-old man consulted his doctor due to symptoms of depression. He did not suffer from any allergy but drinking red wine caused vasomotor rhinitis. Antidepressive treatment with fluoxetine 20 mg daily was initiated which was successful. After three weeks of treatment an itching rash appeared. An adverse drug reaction (ADR) induced by fluoxetine was suspected and fluoxetine treatment was discontinued. The symptoms disappeared with clemastine and betametasone treatment. Since the depressive symptoms returned sertraline medication was initiated. After approximately two weeks of sertraline treatment he noted an intense itching sensation in his scalp after eating a piece of chocolate cake. The itch spread to the arms, abdomen and legs and the patient treated himself with clemastine and the itch disappeared. He now realised that he had eaten a chocolate cake before this episode and remembered that before the first episode he had had a chocolate mousse dessert. He had never had any reaction from eating chocolate before and therefore reported this observation to his doctor. Conclusions This case report suggests that there may be individuals that are very sensitive to increases in serotonin concentrations. Dermal side reactions to SSRI-drugs in these patients may be due to high activity in the serotonergic system at the dermal and epidermo-dermal junctional area rather than a hypersensitivity to the drug molecule itself.	Background The skin contains a system for producing serotonin as well as serotonin receptors. Serotonin can also cause pruritus when injected into the skin. SSRI-drugs increase serotonin concentrations and are known to have pruritus and other dermal side effects e.g. exanthema, purpura, urticaria and pruritus [ 1 ]. In contrast, SSRI-medication has also been used to treat pruritus associated with cholestasis [ 2 ] and polycythemia vera [ 3 ]. In this report we describe a patient who developed pruritus and skin rash from chocolate, but only when he was under SSRI-treatment. The case is presented and we provide a putative biological rationale for the described phenomenon. Case presentation A 46-year-old man consulted his doctor in September 2003 due to depression. He had then experienced symptoms for a few years that had aggravated during the last six to eight months. Using the Montgomery-Åsberg Depression Rate Scale (MADRS) the patient scored 24 points and was diagnosed as having a clinical depression. He did not take any medication and had no regular medical contact. The patient did not have any history of allergy or dermatological diseases. However, he sometimes suffered from vasomotor rhinitis after drinking red wine. The doctor prescribed fluoxetine 20 mg daily as antidepressive treatment. At the revisit three weeks later the patient was very pleased with the fluoxetine treatment and reported that he "had not felt better in 20 years" although he initially had experienced slight nausea and insomnia. A week later, he visited his doctor due to an itching rash that had started the day before. The doctor noted partly confluent urticae on the abdomen, a modest periorbital oedema and red, warm palms and wrists. An ADR induced by fluoxetine was suspected and fluoxetine treatment was discontinued. The symptoms were treated with 2 mg clemastine and 6 mg betametasone orally and disappeared within 48 hours. However, the symptoms of depression returned. Sertraline medication was initiated 10 days after the cessation of fluoxetine treatment since SSRI medication had shown good effect. During the weeks of sertraline treatment no urticarial symptoms appeared. The patient improved in his depression although full recovery was not achieved this time. After approximately two weeks of sertraline treatment he noted an intense itching sensation in his scalp after eating a piece of chocolate cake. The itch spread to the arms, abdomen and legs within a few hours. This time the patient did not seek his doctor but treated himself with clemastine and the itch disappeared during the night. He now remembered that he had had a chocolate mousse dessert before the first episode. Since he had never had any reaction from eating chocolate before, he found this observation so striking that he reported it to his doctor. The patient, himself a scientist, later tried small doses of chocolate and skin rash and itch appeared at an intensity that to him seemed dependent on the "dose" of chocolate ingested. It has been known for 30 years that serotonin can stimulate cutaneous C-fibres [ 4 ], the type of fibres that is also known to transmit itch [ 5 ]. Moreover, serotonin injections into the skin can induce itch [ 6 ] and pruritus is a component in 24% of reported skin reactions to fluoxetine in Sweden, the corresponding figure for sertraline is 15 % [ 1 ]. However, attempts to treat pruritus using 5-HT3-receptor-antagonists have not given clear-cut results [ 6 - 8 ]. The enzymes necessary for conversion of tryptophan to serotonin are expressed in human skin [ 9 ]. In addition, 5-HT2AR are present in one third of unmyelinated axons at the dermal and epidermo-dermal junctional area [ 10 ]. An altered localisation pattern of serotonin receptors 5-HT1AR, 5-HT2AR and 5-HT3R has been reported in contact eczematous skin together with increased serotonin concentrations [ 11 , 12 ] indicating the presence of a serotonin system in the skin that can be altered in pathologic conditions. Moreover, a cross-sensitivity has been reported when skin rash developed after both paroxetine and sertraline medication [ 13 ]. Since these substances are structurally different, one interpretation is that the skin can react to an SSRI-induced increase in serotonin concentrations. In the present case the patient experienced skin symptoms from two different SSRIs. However, these symptoms occurred only when he had eaten chocolate. Chocolate contains serotonin, at concentrations which depend on the type of chocolate [ 14 ]. A concentration of 1.4 – 5 μg / g has been reported in dark chocolate [ 14 ]. The present report suggests an interaction between SSRI-medication and chocolate leading to pruritus and rash. A plausible explanation is that SSRI together with serotonin-containing chocolate has increased serotonin concentration to a level where 5-HT receptors system at the dermal and epidermo-dermal junctional area are affected. Moreover, the patient in this case had previously noted nasal congestion and cough when he was drinking red wine. Red wine can induce release of serotonin from platelets [ 15 ] and from the gut [ 16 ]. Serotonin can induce nasal itch, sneeze and hypersecretion [ 17 , 18 ]. Conclusions Apart from the SSRI – chocolate interaction this patient had another possible sign of sensitivity to serotonin. The present case thus suggests that there may be individuals that are very sensitive to increases in serotonin concentrations. Skin side reactions to SSRI-drugs in these patients may be due to high activity in the serotonergic system system at the dermal and epidermo-dermal junctional area rather than a hypersensitivity to the drug molecule itself. However, the reaction of skin to serotonin from food is poorly studied and further studies are necessary to determine how much alimentary serotonin can increase serum serotonin concentrations and to what extent SSRI-medication affects this process. More knowledge in this field could be of help for physicians who encounter patients with dermal reactions to SSRI-drugs and there might be food and beverages containing serotonin that these patients should avoid. Moreover, possible individual differences in the serotonergic system at the dermal-epidermal junction remain to be studied. What happened to the patient and his depression? Due to poor anti-depressive effect of sertraline, the treatment was altered back to fluoxetine. He is now free from his depression and experiences no rash or oedema-like adverse reactions as long as he is avoiding chocolate. List of abbreviations 5-HT: 5-hydroxytryptamine, ADR: Adverse Drug Reaction, SSRI: serotonin selective reuptake inhibitors Competing interests The author(s) declare that they have no competing interests. Authors' contributions SS first described the case, JC and HM performed literature searches and JC first drafted the manuscript. HM and SK took part in the scientific discussion and in finalising the manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC533866.xml
534112	Mucosal delivery of anti-inflammatory IL-1Ra by sporulating recombinant bacteria	Background Mucosal delivery of therapeutic protein drugs or vaccines is actively investigated, in order to improve bioavailability and avoid side effects associated with systemic administration. Orally administered bacteria, engineered to produce anti-inflammatory cytokines (IL-10, IL-1Ra), have shown localised ameliorating effects in inflammatory gastro-intestinal conditions. However, the possible systemic effects of mucosally delivered recombinant bacteria have not been investigated. Results B. subtilis was engineered to produce the mature human IL-1 receptor antagonist (IL-1Ra). When recombinant B. subtilis was instilled in the distal colon of rats or rabbits, human IL-1Ra was found both in the intestinal lavage and in the serum of treated animals. The IL-1Ra protein in serum was intact and biologically active. IL-1-induced fever, neutrophilia, hypoglycemia and hypoferremia were inhibited in a dose-dependent fashion by intra-colon administration of IL-1Ra-producing B. subtilis . In the mouse, intra-peritoneal treatment with recombinant B. subtilis could inhibit endotoxin-induced shock and death. Instillation in the rabbit colon of another recombinant B. subtilis strain, which releases bioactive human recombinant IL-1β upon autolysis, could induce fever and eventually death, similarly to parenteral administration of high doses of IL-1β. Conclusions A novel system of controlled release of pharmacologically active proteins is described, which exploits bacterial autolysis in a non-permissive environment. Mucosal administration of recombinant B. subtilis causes the release of cytoplasmic recombinant proteins, which can then be found in serum and exert their biological activity in vivo systemically.	Background The use of recombinant proteins as drugs has deeply modified the therapeutic approach to many severe diseases. However, a variety of practical problems limits the use of biotechnological protein drugs. Stability of the active proteins, need for parenteral administration, and high costs of the final purified materials are among the most significant drawbacks. A way of circumventing these issues is represented by the direct administration of recombinant bacteria, acting simultaneously as cell factory and delivery system for pharmacologically active proteins. This approach has been already extensively experimented for the mucosal delivery of vaccine antigens [ 1 , 2 ]. In recent years, the local delivery of therapeutic antibodies [ 3 , 4 ], adjuvant cytokines [ 5 , 6 ], and anti-inflammatory cytokines [ 7 - 9 ] has been successfully attempted with food-grade bacteria ( e.g. , Lactococcus lactis , Streptococcus gordonii ), although limited to the therapy of localised pathologies ( e.g. , inflammatory bowel diseases, IBD, in the gastro-intestinal tract). Among anti-inflammatory strategies, both at systemic and local level, the use of the IL-1 receptor antagonist (IL-1Ra) has received vast attention. IL-1 is a family of cytokines highly active in the modulation of immune amplification and inflammation. The IL-1 family includes two agonist proteins, IL-1α and IL-1β, and one antagonist protein, IL-1Ra. IL-1β is a very potent immunostimulatory and inflammatory cytokine, responsible for initiating and amplifying the host response to invasion. If not properly controlled, IL-1 can cause fever, acute inflammation, tissue destruction, organ failure, and eventually shock and death (reviewed in [ 10 ]). IL-1Ra inhibits IL-1 by acting as a competitive receptor antagonist with no detectable agonist activity, thus representing a natural powerful mechanism to control IL-1-dependent responses and avoid pathological derangement (reviewed in [ 11 , 12 ]). In experimental animal models, IL-1Ra has demonstrated excellent therapeutic effects against acute and chronic inflammatory pathologies, being also effective at high doses in prolonging survival in endotoxic shock [ 11 - 17 ]. In human trials, IL-1Ra has been administered to patients with septic shock, rheumatoid arthritis, graft- versus -host disease, and multiple sclerosis (reviewed in [ 11 , 12 , 16 ]). While only a modest benefit was achieved in patients with septic shock [ 11 , 12 , 16 , 18 ], IL-1Ra had a clear beneficial effect in reducing joint destruction in rheumatoid arthritis [ 11 , 12 , 19 - 21 ]. From the clinical experience with purified recombinant IL-1Ra it became clear that most of the problems of variability of efficacy were due to difficulties in adequate timing and dosage of the drug [ 12 ]. To overcome these problems, gene therapy with adenoviral vectors carrying the IL-1Ra gene has been attempted in experimental animals, yielding promising results in models of type 1 diabetes and ischemic brain damage [ 22 , 23 ]. The clinical application of the gene therapy approach may however meet with difficulties for safety reasons, besides the problems of controlling drug release, concentration, and localisation. Based on previous experience of using recombinant bacteria as in vivo cell factory, here we describe a novel system of local delivery of IL-1Ra, able to achieve systemic effects. The system exploits the ability of certain bacteria (such as Bacillus subtilis ) to undergo autolysis in non-permissive conditions (as it occurs in the mammalian intestine) thereby releasing the cytoplasmic proteins. Intra-colon instillation of B. subtilis expressing recombinant human mature IL-1Ra induces significant serum levels of the recombinant protein in rats and rabbits, and prevents the inflammatory effects of systemic IL-1. Intra-peritoneal administration of recombinant B. subtilis in the mouse could inhibit LPS-induced shock and death. Further experimental evidence with a B. subtilis strain producing human IL-1β demonstrates that this delivery system can be generalised to other recombinant proteins. Results The ability of B. subtilis to generate spores by autolysis of the cell wall, thereby releasing cytoplasmic proteins, was exploited as system for delivery of proteins in vivo . Following a sporulation signal ( e.g. , nutrient depletion) bacteria undergo an autolytic process with release of most of their cellular components and formation of a highly resistant spore containing DNA and few essential proteins (Figure 1A ). As already shown in a previous study [ 24 ], in vitro sporulation of B. subtilis engineered for endocellular expression of human IL-1Ra (strain pSM539) caused the release of large amounts of intact and active recombinant protein within a few hours after the sporulation signal (Figure 1B ). The in vitro release of intracellular recombinant IL-1Ra was equally evident in pSM539 Spo + (normally sporulating) and Spo - bacteria, i.e. genetically modified cells which, in response to the sporulation signal, start the autolysis process but are unable to form a complete spore thus undergoing complete cell destruction (Figure 1C ). Both Spo + and Spo - strains of pSM539 were used in subsequent in vivo experiments with identical results. The IL-1Ra recovered after B. subtilis autolysis in vitro retained full biological activity, with a specific activity of 1.1 × 10 6 inhibitory units (IU)/mg vs. 0.9 × 10 6 IU/mg of reference standard IL-1Ra [ 24 ]. To assess whether the recombinant protein could also be released in vivo by engineered B. subtilis , the bacterial strain engineered with IL-1Ra was administered intra-peritoneally in the mouse, in the small and large intestine of rats, and in the rabbit distal colon. The presence of human IL-1Ra was assessed at different times after administration, both locally and in the serum of treated animals, by Western blotting, ELISA, and BIAcore analysis (Table 1 ). Intact human IL-1Ra was found locally at high levels 3 hours after administration of recombinant bacteria and persisted for several hours. In the serum, intact human IL-1Ra could be found at measurable levels when B. subtilis was administered intra-peritoneally (2/2 mice) or in the colon (5/9 rats, 27/31 rabbits), but not when bacteria were administered in the small intestine (0/5 rats). The delivery of IL-1Ra at the intestinal mucosal level was examined. Live cells of the IL-1Ra-producing pSM539 strain were instilled in the rat distal colon, a non-permissive environment that does not allow B. subtilis vegetative life [ 26 ]. As a control, animals received equal numbers of the pSM214 strain, i.e. B. subtilis cells transformed with the β-lactamase-expressing control plasmid pSM214. Data in Figure 2 show that the presence of intact IL-1Ra can be measured both locally (in the intestinal washings) and in serum for several hours after intra-colonic inoculum of the IL-1Ra-producing B. subtilis strain pSM539, whereas serum of animals receiving B. subtilis pSM214 remained negative. As a control, pSM539 bacteria delivered in the small intestine released detectable amounts of IL-1Ra locally, but no IL-1Ra could be found at the serum level (data not shown; Table 1 ). The serum pharmacokinetic parameters of IL-1Ra released by recombinant B. subtilis , as compared to the purified protein administered intra-colonically, show a few differences (Table 2 ). The C max was higher and the T max quicker for the purified protein, as compared to IL-1Ra released from intra-colonically administered B. subtilis . On the other hand, the AUC/dose was almost identical. Administration of control bacteria pSM214 intra-colonically together with the purified protein did not significantly change the pharmacokinetics parameters of IL-1Ra, except for a slight decrease of the total dose absorbed, indicating that the physical presence of bacteria has little effect on IL-1Ra absorption. It is concluded that engineered B. subtilis delivered intra-colonically releases, conceivably by autolysis, the cytoplasmic recombinant protein, which is subsequently absorbed and can be detected intact at measurable levels in the bloodstream. To verify that IL-1Ra found in serum after B. subtilis administration at the mucosal level is functional, its ability to inhibit IL-1 was evaluated both in vitro and in vivo . In vitro , activity of standard IL-1β was assessed with the classical co-stimulation assay on thymocytes of LPS-unresponsive C3H/HeJ mice. Inhibition by IL-1Ra was evaluated as capacity to decrease IL-1-induced thymocyte proliferation. The presence of biologically active IL-1Ra was measured as inhibition of IL-1β-induced thymocyte proliferation by the IL-1Ra-containing serum of rabbits administered pSM539 intra-colonically. The presence and amount of IL-1Ra was measured by ELISA in the serum of pSM539-treated rabbits and of control pSM214-treated or untreated animals. As shown in Figure 3 , IL-1Ra-containing serum from pSM539-treated rabbit (used at dilutions containing from 0.1 to 10 ng/ml IL-1Ra) was as effective in inhibiting IL-1β activity as the same concentrations of standard purified recombinant IL-1Ra (Figure 3 , left), whereas the same dilutions of serum from pSM214-treated or from untreated animals (devoid of IL-1Ra) did not possess any IL-1-inhibiting activity (Figure 3 , left and right panels). To confirm that the IL-1β-inhibiting activity observed in sera of pSM539 treated animals is indeed due to IL-1Ra, data in the Figure 3 (right) show that the inhibitory capacity of pSM539 serum is significantly decreased or abolished by an antiserum against human IL-1Ra. Thus, it can be concluded that the IL-1Ra present in serum after intra-colonic administration of pSM539 is biologically active. The in vivo efficacy of IL-1Ra released by intra-colonic pSM539 was evaluated in antagonising the effects of parenterally administered IL-1 [ 27 ]. As shown in Figure 4 (upper left), the increase in body temperature induced in rabbits by i.v. administration of 75 ng/kg human IL-1β was significantly reduced by preventive intra-colonic treatment with 2 × 10 9 cells of B. subtilis pSM539. The reduction of IL-1β-induced fever was more pronounced with lower doses of IL-1β (90% reduction of peak fever induced by 50 ng/kg IL-1β), but it was still highly significant when fever was induced by 100 ng/kg IL-1β (>60% reduction of peak fever) (data not shown). To confirm these data, the effect of intra-colonic treatment with IL-1Ra-producing pSM539 was evaluated on other inflammation-related parameters induced by IL-1β, i.e. , granulocytosis and decrease of blood glucose and iron concentrations. As shown in Figure 4 (upper right), the increase in circulating PMN induced in rats by IL-1β i.p. was abrogated by previous intra-colonic administration of IL-1Ra-producing pSM539 but not by the control strain pSM214. Likewise, the IL-1β-induced decrease in the blood levels of iron (Figure 4 , lower left) and glucose (Figure 4 , lower right) was still evident in animals administered control pSM214 bacteria but was significantly reduced by intra-colonic instillation of IL-1Ra-producing pSM539 bacteria. It is inferred that human recombinant IL-1Ra delivered in vivo by intra-colonic administration of engineered B. subtilis is biologically active and able to counteract the systemic inflammatory effects of IL-1. That IL-1Ra delivered by B. subtilis can have an anti-inflammatory protective effect in vivo was shown in a model of shock and death induced by bacterial endotoxin (LPS) in the mouse (Figure 5 ), an acute syndrome in which IL-1β plays a major role [ 14 - 16 , 25 ]. LPS-sensitive C3H/HeOuJ mice receiving recombinant pSM539 bacteria intra-peritoneally 24 hours before administration of a lethal dose of bacterial LPS could survive significantly longer than mice administered the control pSM214 bacteria or PBS, in agreement with previous data on the efficacy of IL-1Ra in inhibiting LPS-induced shock [ 13 - 15 ]. To validate the concept of delivery of bioactive recombinant proteins via the colonic mucosa by means of recombinant B. subtilis , another B. subtilis strain was constructed (pSM261, engineered for production of human mature IL-1β) and administered in vivo to rabbits. As shown in Figure 6 , the intra-colonic administration of 1 × 10 9 live cells of B. subtilis pSM261 induced a significant increase in body temperature, superimposable to that caused by intra-colonic instillation of purified recombinant IL-1β. Furthermore, in agreement with the systemic effects of massive doses of IL-1β administered parenterally [ 25 , 28 ], intra-colonic administration of pSM261 caused shock and death in 9/14 animals (64%). It is concluded that the mucosal delivery of engineered B. subtilis in the large intestine is a suitable system for attaining significant blood levels of bioactive recombinant proteins and systemic effectiveness. Discussion The use of live bacteria is very common in particular in vaccinology, where attenuated or mutant bacteria have been employed for decades as antigen carriers. The advantage of live bacteria relies on their capacity of colonising the host and enter the host organs/tissues with the same modalities as their virulent counterparts, thus eliciting the relevant immune response and immune memory, at variance with killed bacteria or purified bacterial components. Thus, attenuated strains of Salmonella , Listeria monocytogenes , Mycobacterium tuberculosis , Vibrio cholerae are being developed and used as vaccine carriers [ 29 - 31 ]. A further development in the use of live bacteria as antigen carriers in vaccination exploits the technologies of genetic engineering for introducing multiple antigens from different micro-organisms into a single non-virulent bacterial carrier ( e.g. , food-grade lactic acid bacteria), with the possibility of including T- and B-stimulating epitopes from different antigens, and also to engineering into the same carrier adjuvant sequences derived for instance from an immunostimulating cytokine [ 30 - 34 ]. Among bacterial systems developed for antigen delivery in vaccination, some strains of non-pathogenic, food-grade or GRAS (generally regarded as safe) bacteria have been examined for the topical delivery of pharmacologically active protein drugs, after cell engineering with the DNA coding for the protein of interest. This is the case of Lactococcus lactis and of Streptococcus gordonii , which have been engineered to produce recombinant antibodies, adjuvant and anti-inflammatory cytokines, and used to deliver these proteins locally at the mucosal surface after oral administration [ 3 - 9 ]. The goal of these delivery approaches was that of making the recombinant proteins available for therapy of local pathologies or for local effects: antibodies for passive immunotherapy of local infections [ 3 , 4 ], cytokines as adjuvants for mucosal vaccines [ 5 , 6 ], inhibitory cytokines for anti-inflammatory therapy of localised chronic inflammatory diseases (IBD-like pathologies) [ 7 - 9 ]. Although undoubtely promising and susceptible of vast applications, the method of mucosal delivery of therapeutic protein through recombinant bacteria acting as cell factories needs further and deeper investigation. This should include the central issue of safety and contained/controlled release of recombinant micro-organisms [ 8 ], the problem of assessing the mucosal permanence of bacteria (extent and duration of colonisation depending on the changes in the mucosal environment in different conditons of health and nutrition) and the extent of protein release, and the issue of pharmacodynamics of the delivered protein in particular for its systemic effects, beyond the boundaries of the local delivery environment. The delivery system proposed here is not based on the permanence/colonisation capacity of bacteria in the host mucosal surfaces, but it relies on the capacity of sporulating bacteria of releasing intracellular proteins in non-permissive environments. B. subtilis cells engineered to produce human IL-1Ra were able to release the recombinant protein (intact and biologically active) following a sporulation signal in vitro [ 24 ]. This observation could be repeated in vivo , when recombinant B. subtilis cells were inoculated in the intestine of rats or rabbits (a non-permissive environment that does not allow the vegetative life of B. subtilis ). The recombinant protein could be detected locally shortly after administration of bacteria and persisted at measurable levels for several hours. Release and recovery of recombinant IL-1Ra was much more abundant and consistent in the large intestine as compared to the small intestine. Most interestingly, the recombinant protein released from sporulating bacteria delivered in the large intestine was absorbed in the bloodstream at detectable levels, whereas no circulating IL-1Ra could be found after bacterial delivery in the small intestine. IL-1Ra present in the blood was intact, as judged by its molecular mass in Western blotting, and retained full IL-1-inhibiting activity, as judged by its capacity of dose-dependent neutralisation of IL-1β in vitro . The passage of an intact protein from the intestinal lumen to the bloodstream is not a new concept. Indeed, transcytosis has been extensively described in intestinal epithelial cells, and allows transport of intact proteins and macromolecules from the intestinal lumen to the circulation through an endocytic non-degradative pathway in physiological conditions of integrity of the intestinal mucosal barrier [ 35 - 39 ]. This mechanism of transcytotic transport, quantitatively scarce as compared to the degradative pathway of protein absorption, may have a role in physio-pathological passage of antigens, allergens, and toxins. Delivery of IL-1Ra through engineered sporulating bacteria apparently had some pharmacokinetics advantages as compared to the purified protein. Whereas the absorption into the bloodstream was quick after administration of the purified protein (T max at 60 min), IL-1Ra released from intra-colonically administered B. subtilis had a much slower kinetics of absorption (T max 200 min), as expected by the fact that the protein must be released from bacteria before being absorbed. Furthermore, although the C max was decreased for B. subtilis IL-1Ra (136 ng/ml vs. 482 ng/ml for the purified protein; only partially attributable to the higher dosage of the purified protein), the AUC/dose were almost identical. Thus, IL-1Ra delivered intra-colonically by B. subtilis is absorbed into the bloodstream at a slower and more constant rate than the purified protein delivered in the same site, which is absorbed quickly into the bloodstream and rapidly disappears thereafter. Thus, it appears that bacteria do not undergo sporulation all at the same time (which would result in a rapidly appearing and disappearing peak of protein), but release the protein constantly from the moment of administration for about 8 h. This would allow a controlled and sustained circulating level of the protein, thus a more favourable pharmacodynamic profile, with a single administration. The protein selected for in vivo delivery with B. subtilis is the IL-1 receptor antagonist IL-1Ra, a competitive non-activating ligand of the IL-1 receptor with IL-1 inhibitory activity [ 11 , 12 ]. IL-1 is a potent inflammatory cytokine which, in pathological conditions, is responsible of chronicisation of inflammation, tissue destruction, organ failure, hypotensive shock [ 10 ]. Anti-IL-1 strategies have been attempted in acute an chronic inflammatory diseases with the use of recombinant IL-1Ra protein [ 11 , 12 ]. The poor outcome of clinical trials in septic shock has highlighted the problems of a therapy based on the systemic administration of a purified recombinant protein, whose efficacy is hampered by its rapid pharmacokinetics [ 11 , 12 , 18 ]. At present, experimentation of therapeutic IL-1Ra is being targeted to slowly progressive chronic diseases with defined organ/tissue targets ( e.g. , rheumatoid arthritis) [ 40 , 41 ]. To achieve sustained IL-1Ra levels, gene therapy approaches have been attempted with promising results in animal models of experimental arthritis, ischemic brain damage, autoimmune diabetes [ 19 - 23 ]. However, the risk remains of side effects due to the uncontrolled inhibition of the physiologically important IL-1 activity. Indeed, a precise balance between between IL-1β and IL-1Ra should be maintained for achieving proper tissue homeostasis, as shown for the intestinal mucosa [ 42 ]. The drug delivery strategy here described merges the well-known approach of vaccination with live bacteria with that of gene therapy. The delivery of pharmacologically active proteins by live sporulating bacteria, as described here, presents a series of advantages over other similar approaches. At variance with conventional gene therapy, the gene coding for the drug protein is introduced in a bacterial carrier rather than in host cells, a situation that would allow a complete control of its permanence in the body. In a previous study, intragastric or vaginal administration of Streptococcus gordonii engineered to release human IL-1Ra resulted in a prolonged local delivery of the protein, consequent to the capacity of S. gordonii to colonise the mucosal surfaces [ 9 ]. Mucosal delivery of IL-1Ra (by intragastric administration of engineered S. gordonii ) also had a local therapeutic effect in a model of ulcerative colitis [ 9 ]. The delivery system with sporulating bacteria described here differs from that with S. gordonii , as it causes rapid local release of the recombinant protein ( e.g. in the large intestine, where IL-1Ra peaks at 4 h and decreases towards background at 24 h), followed by absorption into the bloodstream. In preliminary experiments in the mouse, IL-1Ra-expressing bacteria were also administered intragastrically or subcutaneously. This achieved appearance of human IL-1Ra in the serum, and systemic effects of inhibition of LPS-induced shock and death (data not shown). This is a new finding, that opens the possibility of exploiting localised bacterial administration ( e.g. at mucosal sites) for systemic drug delivery. The amount of protein released at the mucosal site directly correlates with the number of administered bacteria, since the internal body environment does not sustain bacterial replication but induces sporulation. This allows an exact control of the dose of drug delivered and, based on the pharmacokinetics parameters, of the blood levels that can be reached. The same result could not be easily obtained with S. gordonii , as amount and timing of protein release may be influenced by variation of the colonisation capacity depending on variations of environmental conditions of the host tissues. A problem that should be faced when using recombinant bacteria in vivo for therapy or vaccination is that of safety and contained release of genetically modified organisms (GMO). The use of suicidal genes or the deletion of genes vital for survival outside the host organism have been explored with very promising results [ 8 , 43 ]. The bacterial system proposed here can be modified in the sporulation mechanism for the control of its survival. In preliminary experiments, the recombinant B. subtilis pSM539 strain was engineered in order to inactivate a gene involved in sporulation control. As a consequence, in response to in vitro sporulation signals (adverse environmental conditions) the mutated Spo - strain could regularly initiate the sporulation process, undergoing cell autolysis and release of the cytoplasmic proteins (including the recombinant IL-1Ra), but it was incapable of eventual spore formation and further survival. Likewise, release of the recombinant protein from Spo - in vivo was comparable to that of Spo + bacteria, but spores could never be recovered from intestinal lavage and faeces (data not shown). This suggests that the system can be optimised to full biological containment and environmental safety without altering its delivery properties. Conclusions The novel system of protein drug delivery here proposed links some of the advantages of gene therapy (endogenous production of the relevant protein, targeted delivery) to the possibility of controlled release in terms of timing and protein amount. Exploitation of the mechanism of bacterial autolysis in non-permissive environments allows release of intracellular proteins, including the known amount of the pharmacologically active recombinant protein drug. The release is persistent for several hours, allowing to maintain more constant protein levels in the bloodstream. The system is simple, cheap, and can be developed to full environmental safety ( i.e. , avoiding the risk of release of genetically modified bacteria in the environment). The concept that pharmacologically active proteins released at the colonic mucosal surface can be absorbed and reach the circulation intact and retaining full activity (validated with two proteins with opposite effects, IL-1Ra and IL-1β) opens promising avenues to the use of local delivery for the therapy of systemic diseases. Methods Bacterial strains Engineered B. subtilis strains were constructed as previously described in detail [ 44 , 45 ]. Briefly, cDNA coding for mature human IL-1Ra (encompassing the mutation N91>R), and cDNA coding for mature human IL-1β were cloned between Eco RI and Hind III in pSM214, a B. subtilis plasmid which promotes the synthesis of recombinant products intracellularly, to obtain recombinant plasmids pSM539 (carrying the cDNA for IL-1Ra) and pSM261 (carrying the cDNA for IL-1β). Plasmids were used to transform the B. subtilis SMS118 strain. The pSM539-harbouring B. subtilis strain SMS118(pSM539) could produce 1.0–2.0 mg IL-1Ra/10 9 cells/0.35–0.49 g (wet weight), after conventional culture overnight in 1 liter flasks. The SMS118(pSM261) strain in the same culture conditions produced 0.15–0.25 mg IL-1β/10 9 cells/0.35–0.49 g. As negative control, B. subtilis strain SMS118 was transformed with the pSM214 plasmid, which contains the gene of β-lactamase (conferring resistance to penicillin). All strains were leu - , pyrDI, npr - , apr - . Sporulation-defective (Spo - ) strains were constructed by mutation in the srfA gene, as previously described [ 46 ] and were kindly provided by Dr. G. Grandi (Chiron S.r.l., Siena, Italy). Bacterial preparations Bacteria were grown in LB medium containing 5 mg/l chloramphenicol for 7 h at 37°C under shaking and harvested by centrifugation (3,000 × g, 20 min, 4°C). For sporulation supernatant preparation, 0.25 g wet weight of bacteria (corresponding to 0.5–0.7 × 10 9 cells) were suspended in Difco sporulating medium (bacto beef extract 3 g/l, peptone 5 g/l, NaOH 0.25 mM, MgSO 4 10 mM, KCl 0.1%, MnCl 2 0.1 mM, Ca(NO 3 ) 2 1 mM, FeSO 4 1 mM, pH 6.8) without chloramphenicol and incubated at 35°C with shaking. Aliquots of sporulation supernatant were harvested by centrifugation (14,000 × g, 5 min) at different time points. Following sporulation signals, both Spo + and Spo - bacteria initiate the autolysis process, which ends in cell autolysis with release of cytoplasmic content. However, whereas in Spo + bacteria there is formation of a spore with preservation of strain survival, Spo - bacteria are unable to form a spore thus undergoing complete cell destruction (Figure 1A ). Upon sporulation signals, both Spo + and Spo - bacteria released 100% of intracellular recombinant products in a time-dependent fashion, with maximal relaease between 2 an 8 h (Figure 1C ) [ 24 ]. SDS-PAGE analysis Protein samples were run on 13.5% mini SDS-PAGE according to Lämmli [ 47 ] and stained with Coomassie R-250. The gel was subjected to laser scanning on a Molecular Dynamics Personal Densitometer, and the densitometric analysis was made using Image Quant software (Molecular Dynamics, Sunnyvale, CA). Animals Experimental animals were: female C3H/HeOuJ mice of 10–12 weeks of age (20–25 g) (for all in vivo experiments), female C3H/HeJ mice of 5–8 weeks of age (for thymocyte proliferation), female Sprague-Dawley rats (around 300 g), and female New Zealand rabbits (1.9–2.5 kg). All animals were purchased from Charles River Italia (Calco, Italy) and were housed in standard cages at 22 ± 1°C with 12 h light-12 h dark cycle. Animals received standard diet and tap water ad libitum . In vivo administration of B. subtilis Bacteria were harvested and resuspended in LB medium or sterile PBS. Mice received a single intra-peritoneal injection of 0.2 ml of bacterial suspension in PBS. Rats were fasted overnight before the surgical procedure and maintained under urethane anaesthesia throughout. Bacteria (in LB medium diluted 1:1 in PBS) were instilled in the small intestine (duodenum) with a 22 1/2 G needle, in a volume of 1–10 ml. Two surgical ligatures were applied, one at the beginning of the duodenum immediately below the needle entry puncture (to avoid exit of instilled bacteria), and another one near the ileocecal valve, to limit to the small intestine the transit of bacteria. Intra-colon instillation was performed again with a 22 1/2 G needle in the ceacum immediately below the ileo-caecal valve, in a volume of 5–10 ml. Two surgical ligature were applied just below the needle entry point and at the colon terminal region, to avoid loss of bacteria. Animals were sacrificed by exanguination at different times after treatment, to collect blood and intestinal washings. For intra-colonic administration of bacteria in rabbits, animals were fasted overnight prior to treatment, then lightly restrained in conventional stocks and maintained conscious throughout the experiment. A rounded-tip urethral catheter (Rüsh, Germany) was carefully inserted 10 cm into the distal colon via the anal route and 2 ml of B. subtilis suspension were administered. Serum samples were prepared from blood collected from the rabbit marginal ear vein at different times (0–8 h) after intra-colonic administration of bacteria. In some experiments, animals were sacrificed, to collect the large intestine content (saline washing). Protocols of animal experimentation were reviewed by the institutional ethical board for adherence to ethical guidelines for animal research conduct (Italian D. L.vo 27/01/1992 n. 116 and corresponding EU directive 86/609; policy of refinement, reduction and replacement towards the use of animals for scientific procedures 99/167/EC – Council Decision of 25/1/99), and previously authorised by the Italian Ministry of Health. Detection of human IL-1Ra in animal samples Western blotting: samples were subjected to reducing 15% mini SDS-PAGE and analysed by Western blotting using a polyclonal rabbit serum anti-human IL-1Ra and a goat anti-rabbit IgG secondary antibody conjugated with horseradish peroxidase, as described in detail elsewhere [ 48 ]. Serum samples were filtered on Microcon 100 (MWCO 100,000; Amicon, Beverly, MA) before analysis. ELISA measurement: samples were subjected to quantitative determination of human IL-1Ra using a specific ELISA (Amersham, Little Chalfont, UK), following the manufacturer's instructions. The lower detection limit was 20 pg/ml. Purified human recombinant IL-1Ra was used as standard. Serum samples were filtered on Microcon 100 (Amicon) before analysis. Biosensor measurement: detection of IL-1Ra in serum samples and intestinal washings was confirmed with the biosensor BIAcore™ system (Pharmacia Biosensor AB, Uppsala, Sweden), which allows real time biospecific interaction analysis by means of the optical phenomenon of surface plasmon resonance, as previously described in detail [ 49 ]. The lower detection limit for human IL-1Ra was 2 pg/ml. IL-1-induced thymocyte proliferation The classical assay of co-stimulation of murine thymocyte proliferation was used to evaluate the bioactivity of IL-1 and IL-1Ra. Briefly, thymocytes from 5–8 week-old C3H/HeJ mice (preferentially used because of their LPS unresponsiveness) were cultured at 6 × 10 5 cells/well of Cluster 96 plates (Costar, Cambridge, MA) in 0.2 ml of RPMI-1640 medium (Life Technologies, Paisley, Scotland) supplemented with 2 mM L-glutamine, 25 mM HEPES buffer, 50 μg/ml gentamycin sulfate, 1.25 × 10 -5 M 2-ME (all from Sigma Chemical Co.), 5% fetal bovine serum (Hyclone, Logan, UT) for 72 h in moist air with 5% CO 2 [ 50 ]. The biological activity of IL-1β was assessed as co-stimulation of thymocyte proliferation, by adding to the culture wells a selected amount of human recombinant IL-1β (30–300 pg/ml) [ 51 ] and a suboptimal concentration of purified PHA (1.5 μg/ml; Murex Diagnostics, Dartford, UK). Cells were then pulsed for 18 h with 18.5 kBq/well of [ 3 H]TdR (sp. act. 185 GBq/mmol; Amersham) and their proliferation was measured as radiolabel incorporation with a β-counter. The biological activity of IL-1Ra was evaluated as inhibition of IL-1β-dependent thymocyte proliferation. To this end, cells were stimulated to proliferate (with IL-1β and PHA) in the presence of increasing concentrations of human recombinant IL-1Ra [ 51 ] or serial dilutions of serum from rabbits receiving pSM214 or pSM539 intra-colonically, or from untreated rabbits. The concentration of IL-1Ra in serum of pSM539-treated rabbits was determined by ELISA and serum was added to the cultures after appropriate dilution. Control sera from pSM214-treated or untreated rabbits were used at the same dilutions as IL-1Ra-containing serum. Cell proliferation was then evaluated as radiolabel incorporation as described above. To assess that the effect of IL-1Ra-containing serum was indeed due to IL-1Ra, a polyclonal rabbit antibody against human IL-1Ra [ 48 ] was added to the cultures at a dilution of 1:300, i.e. the dilution previously found to inhibit 50% of the activity of 10 ng/ml IL-1Ra in the thymocyte assay (not shown). IL-1-induced fever Rabbits were lightly restrained in conventional stocks throughout the experiment, and accustomed to the stocks over a period of 2 h, to minimise variations in body temperature. Body temperature was measured by means of a cutaneous thermistor probe (TM-54/S and TMN/S; LSI-Lastem, Settala Premenugo, Italy) placed between the left posterior paw and the abdomen and allowed to stabilise for 2 min. B. subtilis suspensions (2 × 10 9 live cells/rabbit) were instilled in the distal colon 1 h before i.v. administration of 50–100 ng/ml highly purified LPS-free human recombinant IL-1β in pyrogen-free saline through the marginal ear vein. Temperature was recorded every 20 min for 3 h starting from IL-1β administration. In experiments with IL-1β-producing strain pSM261, rabbits received an intra-colonic administration of 1 × 10 9 pSM214 (control) or pSM261 (IL-1β) bacteria, or 250 μg purified human IL-1β. Temperature was recorded up to 22 h after treatment. IL-1-induced neutrophilia, hypoferremia, hypoglycemia Live cells of B. subtilis strains pSM214 and pSM539 were instilled in the distal colon (1 × 10 9 cells/kg), 2 h before administration of IL-1β. Blood samples were drawn 2, 4, 6, 8 and 24 h after intra-peritoneal inoculum of 0.1 μg/kg human recombinant IL-1β. The number of circulating neutrophils was assessed by flow cytometry. The plasma iron concentration was determined colorimetrically with a commercially available kit (Fe; Boehringer Mannheim, Mannheim, Germany). Hypoferremia (60–75% decrease of plasma iron level) was evident from 4 to 24 h after IL-1β inoculum. The blood glucose concentration was measured in serum samples by the glucose/glucose oxidase/peroxidase method with commercially available kits (Glucose GOD Perid; Boehringer Mannheim) or by biosensor detection with devices for diagnostic monitoring (Roche Diagnostics, Milano, Italy). Overlapping results were obtained in rats and rabbits. LPS-induced shock in the mouse LPS-sensitive C3H/HeOuJ mice received an intra-peritoneal inoculum of 0.5 ml PBS alone or containing bacterial suspensions (control pSM214, IL-1Ra-producing pSM539; 3 × 10 6 bacteria/mouse), 24 h before i.p. administration of 15–20 mg/kg of LPS (from E. coli 055:B5; Sigma Chemical Co., St. Louis, MO). LPS inoculum was delayed to 24 h after bacteria administration to avoid interference of pre-inoculum. In fact, preliminary experiments showed that intra-peritoneal inoculum of PBS decreased significantly LPS toxicity when administered at shorter times before LPS (data not shown). Mice were observed for 7 days after LPS administration and deaths recorded. Statistical analysis Results are presented as mean ± SEM. Statistical significance was assesed by two-tailed Student's t test. Comparison of survival curves was performed by the χ 2 test. Calculation of percentiles was performed by survival analysis. All calculations were performed with the Stratgraphics Plus 5 programme (Manugistics, Inc., Rockville, MD). List of abbreviations IL, interleukin; IL-1, interleukin-1; IL-1Ra, interleukin-1 receptor antagonist; IBD, inflammatory bowel disease; LPS, bacterial lipopolysaccharide, AUC, area under the curve; PMN, polymorphonuclear leukocytes; GRAS, generally regarded as safe; GMO, genetically modified organisms. Authors' contributions SP carried out the in vivo and pharmacokinetics studies in rats and rabbits, and performed the statistical analysis. PB designed and performed the bioactivity studies. PR designed and performed the microbiological and biochemical work. DB coordinated the bioactivity studies, organised the data, and wrote the manuscript. AT designed and coordinated the entire study. All authors read and approved the final manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC534112.xml
545066	Mycophenolate mofetil modulates adhesion receptors of the beta1 integrin family on tumor cells: impact on tumor recurrence and malignancy	Background Tumor development remains one of the major obstacles following organ transplantation. Immunosuppressive drugs such as cyclosporine and tacrolimus directly contribute to enhanced malignancy, whereas the influence of the novel compound mycophenolate mofetil (MMF) on tumor cell dissemination has not been explored. We therefore investigated the adhesion capacity of colon, pancreas, prostate and kidney carcinoma cell lines to endothelium, as well as their beta1 integrin expression profile before and after MMF treatment. Methods Tumor cell adhesion to endothelial cell monolayers was evaluated in the presence of 0.1 and 1 μM MMF and compared to unstimulated controls. beta1 integrin analysis included alpha1beta1 (CD49a), alpha2beta1 (CD49b), alpha3beta1 (CD49c), alpha4beta1 (CD49d), alpha5beta1 (CD49e), and alpha6beta1 (CD49f) receptors, and was carried out by reverse transcriptase-polymerase chain reaction, confocal microscopy and flow cytometry. Results Adhesion of the colon carcinoma cell line HT-29 was strongly reduced in the presence of 0.1 μM MMF. This effect was accompanied by down-regulation of alpha3beta1 and alpha6beta1 surface expression and of alpha3beta1 and alpha6beta1 coding mRNA. Adhesion of the prostate tumor cell line DU-145 was blocked dose-dependently by MMF. In contrast to MMF's effects on HT-29 cells, MMF dose-dependently up-regulated alpha1beta1, alpha2beta1, alpha3beta1, and alpha5beta1 on DU-145 tumor cell membranes. Conclusion We conclude that MMF possesses distinct anti-tumoral properties, particularly in colon and prostate carcinoma cells. Adhesion blockage of HT-29 cells was due to the loss of alpha3beta1 and alpha6beta1 surface expression, which might contribute to a reduced invasive behaviour of this tumor entity. The enhancement of integrin beta1 subtypes observed in DU-145 cells possibly causes re-differentiation towards a low-invasive phenotype.	Background With the improved long-term outcome of allograft recipients in the cyclosporine or tacrolimus era, malignant tumors have become increasingly important. Malignant tumours develop in 15–20% of graft recipients after 10 years, and thus contribute substantially to the morbidity and mortality of these patients [ 1 ]. Malignancies can develop in three ways: de-novo occurrence in the recipient, recurrent malignancy in the recipient or transmission of malignancy from the donor. In all cases, the post-transplant treatment regimen and the level of immunosuppression are high risk factors due to the long-term modification of the immune system. During the last years, the novel immunosuppressive drug mycophenolate mofetil (MMF) has been introduced into the clinical protocol to overcome severe side effects associated with cyclosporine or tacrolimus. Meanwhile, it has become part of the immunosuppressive regimen after liver, kidney or heart transplantation [ 2 ]. Still, the influence of MMF on tumor recurrence or de novo malignancy has not been explored. MMF effects are based on the inhibition of inosine monophosphate dehydrogenase (IMPDH) and prevention of guanosine monophosphate synthesis from inosine monophosphate, a rate-limiting step in the purine biosynthesis in lymphocytes. Consequently, MMF blocks the proliferation and clonal expansion of T and B lymphocytes, and prevents the generation of cytotoxic T cells, as well as other effector T cells [ 3 ]. Additional mechanisms may also contribute to the efficacy of MMF in preventing allograft rejection. By depleting guanosine nucleotides, MMF suppresses glycosylation and the expression of some adhesion molecules, thereby decreasing the recruitment of lymphocytes and monocytes into sites of inflammation and graft rejection [ 3 ]. Immunoprecipitation studies have shown that one of the glycoproteins affected is the lymphocytic alpha4beta1 integrin, the ligand for VCAM-1 on activated endothelial cells. Further experiments have revealed inhibition of the integrin LFA-1, the counter-receptor of ICAM-1, after MMF administration [ 4 , 5 ]. The integrins constitute a family of transmembrane receptor proteins composed of heterodimeric complexes of noncovalently linked alpha and beta chains. Integrins function in cell-to-cell and cell-to-extracellular matrix (ECM) adhesive interactions and transduce signals from the ECM to the cell interior and vice versa. For various types of cancers, different changes in integrin expression are closely associated with tumor growth and metastasis. Based on the knowledge that MMF modulates integrin expression, we postulated that MMF might not only suppress leukocyte recruitment to the donor graft, but also prevent integrin-dependent tumor dissemination. To explore how far MMF might serve as a metastasis-blocking agent, we investigated the beta1 integrin subunit expression pattern of colon, kidney, pancreas and prostate tumor cells before and after MMF treatment, as well as MMF effects on tumor cell adhesion to human endothelium in vitro. The present study indicates that MMF possesses anti-tumoral properties particularly to colon and prostate carcinoma cells. Alterations of the beta1 integrin profile are responsible for blocking tumor cell adhesion to vascular endothelium. Methods Cell cultures Kidney carcinoma Caki I cells, pancreatic carcinoma DanG cells and colonic adenocarcinoma HT-29 cells G were obtained from the tumor cell bank of Johannes Gutenberg University, Mainz, Germany. Prostate carcinoma DU-145 cells were purchased from DSMZ (Braunschweig, Germany). Tumor cells were grown and subcultured in RPMI1640 medium (Seromed, Berlin, Germany) supplemented with 10% FCS, 100 IU/ml penicillin and 100 μg/ml streptomycin at 37°C in a humidified, 5% CO 2 incubator. Endothelial cells (HUVEC) were isolated from human umbilical veins and harvested by enzymatic treatment with chymotrypsin. HUVEC were grown in Medium 199 (Biozol, Munich, Germany), 10% fetal calf serum (FCS; Gibco, Karlsruhe, Germany), 10% pooled human serum (Blood Bank of The German Red Cross, Frankfurt am Main, Germany), 20 μg/ml endothelial cell growth factor (Boehringer, Mannheim, Germany), 0.1% heparin (Roche, Basel, Switzerland), 100 ng/ml gentamycin (Gibco) and 2% 1 M HEPES-buffer (Seromed, Berlin, Germany). To control the purity of HUVEC cultures, cells were stained with fluorescein isothiocyanate (FITC)- labelled monoclonal antibody against Factor VIII-associated antigen (Von Willebrand factor; clone F8/86; Dako, Hamburg, Germany) and analyzed microscopically or by FACscan (Becton Dickinson, Heidelberg, Germany; FL-1H (log) channel histogram analysis; 1 × 10 4 cells/scan). Cell cultures with a purity > 95% were serially passaged. Subcultures from passages 2–4 were selected for experimental use. Mycophenolate mofetil (MMF) Tumor cells were pretreated with MMF (Roche Bioscience, Grenzach-Wyhlen, Germany) (0.1 μM, 1 μM). Before adding the MMF-treated tumor cells to HUVEC (monolayer adhesion assay), cell cultures were washed to remove MMF from the medium. Results were compared to untreated controls. Viability of tumor cells in presence of MMF was assessed by propidium iodide dsDNA-intercalation or quantitative fluorescence analysis of enzyme-catalyzed fluorescein-diacetate metabolism. Monolayer adhesion assay HUVEC were transferred to six-well multiplates (Falcon Primaria; Becton Dickinson, Heidelberg, Germany) in complete HUVEC-medium. When confluency was reached, 0.5 × 10 6 tumor cells of each entity/well were carefully added to the HUVEC monolayer for 60 min. Subsequently, non-adherent tumor cells were washed off using warmed (37°C) Medium 199. The adherent cells were fixed with 1% glutaraldehyde and counted in five different fields (5 × 0.25 mm 2 ) using a phase contrast microscope (20 × objective) to calculate the mean cellular adhesion rate. Evaluation of integrin surface expression Tumor cells were washed in blocking solution (PBS, 0.5% BSA) and then incubated for 60 min at 4°C with the FITC-conjugated monoclonal antibody anti-alpha2beta1 (Becton Dickinson; clone AK-7), anti-alpha4beta1 (Cymbus Biotechnology, Hofheim, Germany; clone HP2I1), anti-alpha5beta1 (Cymbus Biotechnology; clone SAM-1), anti-alpha6beta1 (Becton Dickinson; clone GOH3), or with the PE-conjugated monoclonal antibody anti-alpha1beta1 (Becton Dickinson; clone SR84), or anti-alpha3beta1 (Becton Dickinson; clone C3II1). Integrin expression of tumor cells was then measured using a FACscan (Becton Dickinson; FL-1H (log) channel histogram analysis; 1 × 10 4 cells/scan) and expressed as mean fluorescence units (MFU). A mouse IgG1-FITC was used as an isotype control for FITC conjugated antibodies. To evaluate background staining of PE conjugated antibodies, goat anti mouse IgG-PE was used (all: Cymbus Biotechnology). To analyze integrin beta1 distribution on the cell membrane, tumor cells were transferred to round cover slips (pretreated with 2% 3-aminopropyl-triethoxysilan) placed in a 24 well multiplate. Upon reaching confluency, cell cultures were washed and fixed in cold (-20°C) methanol/acetone (60/40 v/v). Subsequently, cells were incubated for 60 min with unconjugated anti-integrin monoclonal antibodies. Indocarbocyanine (Cy 3™; Dianova; working dilution: 1:50) conjugated goat-anti-mouse IgG was then added as the secondary antibody. To prevent photobleaching of the fluorescent dye, cover glasses with stained cells were taken out of the wells and the residual liquid was removed. These were then embedded in an antifade reagent / mounting medium mixture (ProLong™ Antifade Kit, MoBiTec, Göttingen, Germany) and mounted on slides. The slides were viewed using a confocal laser scanning microscope (LSM 10; Zeiss, Jena, Germany) with a plan-neofluar ×100 / 1.3 oil immersion objective. mRNA expression of beta1 integrins mRNA expression of beta1 integrins was evaluated by reverse transcriptase-polymerase chain reaction (RT-PCR). Tumor cells were seeded in 50 ml culture flasks (25 cm 2 growth area; Falcon Primaria, Becton Dickinson) and cultured with or without MMF. Total RNA was extracted by using RNeasy kit (Qiagen, Hilden, Germany) and RNA samples were then treated with 80 U/ml of Rnase-free Dnase I (Boehringer Mannheim, Mannheim, Germany) for 60 min at 37°C, to eliminate amplifiable contaminating genomic DNA. Subsequently, samples were incubated for 10 min at 65°C to inactivate Dnase. Complementary DNA was synthesized from 1 μg of total RNA per sample with a 60 min incubation at 42°C, using the Moloney murine leukaemia virus reverse transcriptase (Invitrogen, Karlsruhe, Germany) and oligo-(dT) priming (Boehringer Mannheim). Amplification was carried out using gene specific primers and Platinum-Taq polymerase (Invitrogen) in a Mastercycler Gradient thermocycler (Eppendorf, Hamburg, Germany). Reactions were performed in the presence of 0.5 μl cDNA, with an initial incubation step at 95°C for 2 min. Cycling conditions consisted of denaturation at 95°C for 30 sec, annealing at 60°C for 30 sec and extension at 72°C for 30 sec over a total of 30 cycles. The reaction was completed by another 10 min incubation step at 72°C. The specific sequences for sense and anti-sense primers are shown in table 1 . The PCR products were subjected to electrophoresis in 1.5% agarose gel and visualized by ethidium bromide. Statistical analysis All studies were performed 3–6 times. Statistical significance was investigated by the Wilcoxon signed rank test showing two-sided probabilities and using normal approximation. Differences were considered statistically significant at a p value less than 0.05. Results MMF modulates tumor cell adhesion to HUVEC The 60 min adhesion rates of tumor cells were calculated at 22.5 ± 4.1 DanG cells/0.25 mm 2 , 39,8 ± 10.5 DU-145 cells/0.25 mm 2 , 55,3 ± 11.7 Caki I cells/0.25 mm 2 , or 80,5 ± 17.2 HT-29 cells/0.25 mm 2 . MMF differentially modulated the adhesive capacity of the tumor cells which was strongly dependent on the drug concentration and the cell line used (figure 1 ). Adhesion of DanG cells was weakly reduced by 1 μM MMF. A modest down-regulating effect was seen on Caki I cells at a MMF concentration of 0.1 μM. Strong and significant adhesion blockade was achieved when HT-29 cells were treated with 0.1 μM MMF (p = 0.0079). This effect was reverted at a dosage of 1 μM. Furthermore, MMF dose-dependently and significantly reduced the adhesive capacity of DU-145 cells with a maximum effect at 1 μM (p = 0.0079). In all experiments, cell viability was not impaired by MMF. beta1 integrin expression pattern Figures 2 and 3 depict the integrin beta1 surface expression pattern on untreated tumor cell cultures. Figure 2 is related to FITC-labelled antibodies, figure 3 to PE-labelled antibodies. Each tumor entity was characterized by a specific integrin pattern. Integrins were expressed in the following order (MFU ± SD; n = 4): DanG: alpha3beta1 (455.6 ± 71.0) > alpha2beta1 (175.7 ± 24.3) > alpha6beta1 (54.8 ± 9.2) > alpha1beta1 (27.3 ± 3.2). Caki I: alpha3beta1 (601.3 ± 82.0) > alpha1beta1 (139.2 ± 24.0) > alpha5beta1 (22.9 ± 4.2). HT-29: alpha3beta1 (202.0 ± 33.8) > alpha6beta1 (119.6 ± 14.5) > alpha2beta1 (69.2 ± 9.0) > alpha1beta1 (25.4 ± 3.4). DU-145: alpha3beta1 (942.5 ± 112.9) > alpha2beta1 (95.4 ± 12.4) > alpha5beta1 (69.1 ± 8.9) > alpha6beta1 (50.9 ± 7.2) > alpha1beta1 (31.4 ± 4.8). Mean IgG1-FITC isotype control was 8.6 ± 1.8 MFU, mean IgG1-PE isotype control was 9.9 ± 1.7 MFU. Analysis of the mRNA expression level confirmed the flow cytometry data (see below). The distribution pattern of those integrins which were predominantly expressed on the respective tumor cell line was further explored by confocal microscopy (figure 4 ). alpha2beta1 (DanG cells), alpha3beta1 (Caki I, DanG HT-29 cells), and alpha6beta1 integrins (HT-29 cells) were distributed homogenously among the cell surface. In contrast, alpha1beta1 integrins on Caki I cells accumulated mainly at the sites of cell-cell-contacts. MMF modulates beta1 integrin surface expression MMF evoked distinct alterations of the beta1 integrin expression pattern (figure 5 ). MMF only slightly changed alpha1beta1, alpha3beta1, and alpha5beta1 integrin surface levels on Caki I cells (figure 5A ), and weakly down-regulated alpha2beta1 on DanG cells when applied at 1 μM (figure 5B ). However, 0.1 μM MMF strongly and significantly diminished alpha3beta1 (p = 0.0022) and alpha6beta1 integrins (p = 0.035) on HT-29 cells (figure 5C ). This effect was reverted at concentrations of 1 μM MMF. The alpha3beta1 receptor became even slightly enhanced, compared to control values. alpha1beta1, alpha2beta1, alpha3beta1, alpha5beta1 on DU-145 cells were up-regulated significantly by MMF in a dose-dependent fashion, whereby strongest effects were seen on alpha2beta1 surface level in the presence of 1 μM MMF (>70% fluorescence enhancement, compared to non-treated controls; p = 0.0022; figure 5D ). Influence of MMF on beta1 integrin coding mRNA To allow a clear interpretation of the strong effects of MMF on adhesion and beta1 integrin surface expression of HT-29 and DU-145 cells, MMF evoked alterations of gene activity was also evaluated in these cell lines (figure 6 ). Control experiments using non-treated HT-29 cells revealed high alpha2beta1, alpha3beta1, alpha6beta1 mRNA expression level (figure 6A ). Application of 0.1 μM MMF induced down-regulation of alpha3beta1 and alpha6beta1 coding mRNA, which paralleled MMF's influence on receptor surface expression. The effect was reverted at a dosage of 1 μM. alpha3beta1 and alpha6beta1 coding mRNA became even slightly enhanced, compared to control experiments. With respect to the prostate tumor cell line DU-145, alpha1beta1, alpha2beta1, alpha3beta1 and alpha6beta1 coding mRNA was clearly detected in untreated cell cultures. Both, 0.1 μM and 1 μM MMF reduced mRNA of beta1 integrin subtypes, although the effect was more pronounced in the presence of 0.1 μM MMF (figure 6B ). Discussion Although MMF has become part of the standard regimen after organ transplantation its impact on tumor development and dissemination is still not clear. Our adhesion experiments demonstrate that MMF down-regulates binding of tumor cells to endothelium, which might argue for anti-tumoral properties of this compound. Notably, HT-29 and DU-145 cells responded well to MMF (-70% adhesion reduction), while Caki I and DanG tumor cells were influenced only modestly. From a clinical viewpoint, distinct adhesion-blocking properties of MMF might be limited to colon and prostate carcinoma cells. Interestingly, HT-29 cells were more susceptible to MMF than DU-145 cells: 0.1 μM MMF was sufficient to significantly diminish adhesion of HT-29 cells, whereas 1 μM MMF was necessary to evoke maximum effects on DU-145 cells. The different sensitivity of the tumor cell lines to MMF might be caused by an unequal metabolic activity, coupled with variable IMPDH levels. Recent data have shown that the level of expression of IMPDH mRNA and protein differ among several cell lines [ 6 ], and that IMPDH is selectively up-regulated in neoplastic and replicating cells [ 7 ]. Although this has not yet been proven, MMF might be more effective in rapidly proliferating tumor cells than in tumors with a lower replicating activity. In this context, the average doubling times of HT-29 and DU-145 cultures during their exponential growth phase were calculated to be 13–16 h or 22 h, respectively [ 8 - 10 ], whereas the mean population doubling times of renal or pancreatic carcinoma cell lines ranged between 24–104 h or 16–40 h, respectively [ 11 - 14 ]. It should also be considered that MMF might switch on/off different intracellular signaling cascades in colon versus prostate tumor cells. Indeed, adhesion blockade of HT-29 cells was accompanied by reduced alpha3beta1 and alpha6beta1 surface expression, while adhesion blockade of DU-145 cells was accompanied by a dose-dependent up-regulation of integrins alpha1beta1, alpha2beta1, alpha3beta1, alpha5beta1 on the cell membrane. Studies on integrin receptors presented evidence that beta1 integrin expression by colon carcinoma cells qualifies these cells to successfully adhere to secondary sites. Recent experiments have demonstrated that colon cancer cells adhere to endothelial cells via beta1 integrins and that addition of beta1 integrin blocking antibodies reduces tumor cell adhesion [ 15 , 16 ]. Based on a murine spleen injection-liver metastasis protocol, the alpha3beta1 integrin subtype was identified to predominantly facilitate the metastatic activity of colon cancer cells [ 17 ]. A converse scenario might be created during prostate carcinogenesis, as levels of beta1 integrins have been found reduced in neoplastic versus normal prostate tissue [ 18 , 19 ], and in malignant versus non-tumorigenic prostate cell lines [ 20 ]. An in vitro cell culture model revealed that TGF-beta stimulates the expression of alpha2beta1 integrin on prostate cancer cell lines and concomitantly reduces tumor cell adhesion to human bone marrow endothelium [ 21 ]. Down-regulation in the expression of the alpha3beta1 integrins may also allow prostate tumor cells to become more invasive and lead to an increased propensity for metastasis: When human alpha3beta1 high and alpha3beta1 low expressing prostate carcinoma cells were injected into immunocompromised SCID mice, only those cells with a drastically reduced integrin level were found to form tumors at the primary sites and to be highly invasive and metastatic [ 22 ]. This is in context with our data demonstrating beta1 integrin elevation on DU-145 prostate tumor cells in the context with diminished adhesion behaviour. When discussing the relevance of integrins in tumor recurrence and malignancy, we should keep in mind that integrin receptors serve as mechanistic binding as well as differentiation triggering elements. Therefore, up-regulation of the same integrin type might either lead to enhanced cell adhesion by coupling the receptor to its ligand, or to a reduced cell adhesion by activating integrin driven differentiation signals. Based upon our in vitro assay, we conclude that MMF blocks adhesion of colon and prostate carcinoma cells by two different mechanisms: a) Loss of alpha3beta1 and alpha6beta1 surface expression directly contributes to the reduced adhesive behaviour of HT-29 cells, b) Enhancement of integrin beta1 subtypes might cause re-differentiation of DU-145 cells towards a low-adhesive phenotype. However, it still remains to be determined if MMF indeed acts as a differentiation inducing drug in prostate tumor cells. Beside the hypothesis that beta1 upregulation might activate differentiation inducing signals, selective inhibition of tumor-promoting pathways should also taken into consideration. Presumably, down-regulation of alpha3beta1 and alpha6beta1 surface expression on HT-29 tumor cells might be caused by inhibition of receptor glycosylation and/or receptor de novo synthesis. The latter hypothesis seems to be more likely because MMF's effects at the cell surface were also observed at the mRNA level. This was not the case with DU-145 cells where MMF evoked up-regulation of membranous beta1 integrins was not paralleled by similar modifications of the beta1 integrin coding mRNA. There is still no clear concept why MMF causes integrin up-regulation in one tumor entity but down-regulation in another entity, both coupled with reduced tumor cell adhesiveness. Presumably, HT-29 and DU-145 tumor cells might be equipped with different enzyme systems, the intracellular signaling cascade might be activated differentially in colon versus prostate tumor cells, or sensitivity of specific pathways to MMF might differ between both tumor types. Speculatively, alterations of post-translational events might change the receptor surface presentation in prostate carcinoma cells. Elegant experiments by Liang and coworkers demonstrated that over-expression of alpha5beta1 or beta1 integrin induced the decrease of protein kinase B (PKB) phosphorylation and subsequent accumulation of cyclin-dependent kinase inhibitor p21 [ 23 ]. A yeast-based two-hybrid system was employed which identified IMPDH as specifically interacting with PKB [ 24 ]. Furthermore, MMF treatment significantly increased p21 proteins, which could be reversed by the simultaneous addition of guanine or guanosine [ 25 , 26 ]. Hypothetically, p21 may act as an MMF triggered upstream signal (via PKB?), which contributes to enhanced beta1 integrin surface expression. Conclusions The present study indicates that MMF possesses anti-tumoral properties particularly to colon and prostate carcinoma cells. Alterations of the beta1 integrin profile are responsible for blocking tumor cell adhesion to vascular endothelium. MMF might also act on further adhesion proteins which are relevant for tumor recurrence and dissemination. An in vitro study published recently refers to the sLeX-selectin pathway targeted by MMF [ 27 ]. CD44 glycoproteins as well as receptors of the cadherin family might also be modulated under MMF-based immunosuppressive regimen. From a clinical viewpoint, further studies must be undertaken which evaluate the tumor recurrence rate and classify the tumor type in MMF versus non-MMF treated transplant patients. Competing interests The author(s) declare that they have no competing interests. Authors' contributions TE performed parts of the in vitro studies, contributed toward the design of the study and drafted the manuscript. JM carried out confocal microscopy, BR and IN performed FACS-analyses. IM designed PCR primers and carried out the PCR studies. WDB contributed to the manuscript design and finalisation. DJ participated in the conception and design of the study. RAB carried out the adhesion assays, participated in the conception and design of the study and its coordination. All authors read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC545066.xml
509250	Evaluation of reporting timeliness of public health surveillance systems for infectious diseases	Background Timeliness is a key performance measure of public health surveillance systems. Timeliness can vary by disease, intended use of the data, and public health system level. Studies were reviewed to describe methods used to evaluate timeliness and the reporting timeliness of National Notifiable Diseases Surveillance System (NNDSS) data was evaluated to determine if this system could support timely notification and state response to multistate outbreaks. Methods Published papers that quantitatively measured timeliness of infectious disease surveillance systems operating in the U.S. were reviewed. Median reporting timeliness lags were computed for selected nationally notifiable infectious diseases based on a state-assigned week number and various date types. The percentage of cases reported within the estimated incubation periods for each disease was also computed. Results Few studies have published quantitative measures of reporting timeliness; these studies do not evaluate timeliness in a standard manner. When timeliness of NNDSS data was evaluated, the median national reporting delay, based on date of disease onset, ranged from 12 days for meningococcal disease to 40 days for pertussis. Diseases with the longer incubation periods tended to have a higher percentage of cases reported within its incubation period. For acute hepatitis A virus infection, which had the longest incubation period of the diseases studied, more than 60% of cases were reported within one incubation period for each date type reported. For cryptosporidiosis, Escherichia coli O157:H7 infection, meningococcal disease, salmonellosis, and shigellosis, less than 40% of cases were reported within one incubation period for each reported date type. Conclusion Published evaluations of infectious disease surveillance reporting timeliness are few in number and are not comparable. A more standardized approach for evaluating and describing surveillance system timeliness should be considered; a recommended methodology is presented. Our analysis of NNDSS reporting timeliness indicated that among the conditions evaluated (except for acute hepatitis A infection), the long reporting lag and the variability across states limits the usefulness of NNDSS data and aberration detection analysis of those data for identification of and timely response to multistate outbreaks. Further evaluation of the factors that contribute to NNDSS reporting timeliness is warranted.	Background Public health surveillance is defined as the "ongoing systematic collection, analysis, and interpretation of data essential to the planning, implementation, and evaluation of public health practice, closely integrated with the timely dissemination of these data to those who need to know"[ 1 ]. Reasons for conducting public health surveillance can include the need to assess the health status of a population, establish public health priorities, and reduce the burden of disease in a population by appropriately targeting effective disease prevention and control activities [ 2 ]. Timeliness is a key surveillance system metric and should be periodically evaluated [ 3 , 4 ] because it can reflect the time delay between any number of response steps in the public health surveillance process. Surveillance system timeliness depends on a number of factors and its assessment should include a consideration of how the data will be used and the nature of the condition under surveillance (e.g., for infectious diseases, this includes the communicability of the disease) [ 3 ]. If the data are to be used to implement immediate disease control and prevention activities for infectious diseases that are acute, severe, and highly transmissible, timeliness is critical. Timeliness requirements for a surveillance system might vary by different levels of public health system (e.g., local, state, or national), on the basis of the intended uses of the surveillance data at that level (Table 1 ). For example, timely data are needed within a state for identifying cases or clusters of disease that will prompt an immediate public health response. Timely national surveillance data aggregated from a number of jurisdictions may be used for identifying multistate outbreaks or disease clusters and enable the federal public health system to assist the states in performing and coordinating their prevention and control activities. In reportable disease surveillance systems, health care providers and diagnostic laboratories usually report information regarding persons with notifiable conditions to the local public health system. Then, reporting proceeds in a hierarchical fashion to the state and then to the national level. Health care provider and public health system actions at each successive level of the reporting hierarchy contribute to reporting timeliness delays at the national level. Table 1 Potential uses of infectious disease surveillance data, by level of the public health system Intended Uses Used at which level(s) of the public health system?* Identify individual cases or clusters in a jurisdiction to prompt intervention or prevention activities Local, State (National) Identify multi-state disease outbreaks or clusters. State, National Monitor trends to assess the public health impact of the condition under surveillance. State, National (Local) Demonstrate the need for public health intervention programs and resources, as well as allocate resources. State, National (Local) Monitor effectiveness of prevention, control, and intervention activities. State, National (Local) Formulate hypotheses for further study. National (State) Public health system level in parentheses represents secondary use of the data for that purpose. State and national surveillance processes Before data can be used for public health action, health-related data must be collected by the public health system, analyzed, and disseminated to those responsible for taking action (Figure 1 ). Within a state (Steps 1–7), the public health system can use surveillance data for a number of purposes, including outbreak detection and intervention planning and implementation (Table 1 ). The number and sequence of actions a state conducts before reporting data to the national public health system might vary by state, depending on state policies and protocols (Figure 1 ). For example, for nationally notifiable infectious disease reporting, CDC recommends that states report as soon as they first receive information about a suspect, probable, or confirmed case. However, some states only report confirmed cases, which usually requires laboratory confirmation, and decreases reporting timeliness at the national level. Figure 1 Sequence of actions needed to gather and use health-related information for public health purposes Each week, states and the U.S. territories report case information on persons suspected of having or diagnosed with a nationally notifiable infectious disease to the Nationally Notifiable Diseases Surveillance System (NNDSS), maintained by the Centers for Disease Control and Prevention (CDC) [ 5 ]. A nationally notifiable disease is one for which "regular, frequent, and timely information regarding individual cases is considered necessary for prevention and control of the disease" [ 6 ]. At the national level, NNDSS data are used for monitoring trends, program planning, evaluation, policy development, research, and monitoring the effectiveness of prevention and control activities. Although NNDSS reporting timeliness for these long-range goals and objectives is not critical, the threat of terrorism prompted consideration of whether NNDSS could be enhanced in the future to support public health response for either naturally occurring diseases or terrorism preparedness and response efforts. Therefore, the timeliness of NNDSS data was evaluated to determine if NNDSS could support timely notification and state response to multistate outbreaks. To provide a context for the evaluation of NNDSS timeliness, published studies reporting timeliness measures for infectious disease surveillance systems in the United States were reviewed. Methods Literature review Infectious disease surveillance evaluation studies reporting timeliness measures that were published between January 1970 and March 2003 in biomedical and public health literature were reviewed. English-language papers were identified by using the Medline database (U.S. National Library of Medicine). The search strategy used various combinations of the following key words "timeliness," "reporting delay," "time delay," "lag time," "disease surveillance," "disease outbreaks," "communicable diseases," and "infectious diseases." Reference lists of the studies identified through the Medline search and studies citing CDC's surveillance evaluation guidelines were also reviewed [ 3 , 7 ] Reports were included if they evaluated a public health surveillance system operating in the United States and provided a quantitative estimate of disease-specific timeliness (e.g., interval in days). Studies without quantitative timeliness estimates or that reported a quantitative estimate for a group of infectious diseases (versus a disease-specific estimate) were excluded. In addition, studies describing the timeliness of syndromic surveillance systems were excluded. Information abstracted for the review included the disease(s) under surveillance, the geographic area and time period studied, the purpose of the surveillance evaluation, the surveillance time interval measured, the surveillance processes or actions (steps in Figure 1 ) covered within the measured time interval, the timeliness measure, and the study's assessment of whether surveillance data timeliness met the surveillance goals. NNDSS timeliness Information available for assessing NNDSS reporting timeliness includes the Morbidity and Mortality Weekly Report [ MMWR ] week number the state assigns to each case and one of the following earliest known dates associated with the incidence of this disease (earliest known date) from the following list of hierarchical date types: onset date, diagnosis date, date of laboratory result, or date of first report to the community health system. National reporting delay was calculated as the difference in days between the midpoint of the MMWR week and the earliest known date reported in association with the case. This time interval reflects various state-specific surveillance intervals in the surveillance process that occur between the occurrence of a health event and the reporting of that health event to NNDSS, but at a minimum it includes Intervals 1–4 (Figure 1 ). National median reporting timeliness was calculated overall for the years 1999–2001, for each disease in our study, by date type and state, and across all states. Median reporting delay was calculated using Proc means in SAS version 8 software for Windows (SAS Institute, Inc., Cary, North Carolina). To assess whether analysis of NNDSS data could support the timely identification of multistate outbreaks at the national level, the percentage of NNDSS cases reports reported within one to two incubation periods for each of the diseases was determined. Incubation periods were used as a surrogate measure for period of communicability which is critical to consider when implementing effective, disease-specific prevention and control measures. For this analysis, estimated incubation periods were used for the seven nationally notifiable infectious diseases selected for this study: 7 days for cryptosporidiosis, 4 days for Escherichia coli O157:H7 ( E. coli ), 30 days for acute hepatitis A virus infection, 4 days for meningococcal disease, 20 days for pertussis, 1.5 days for salmonellosis, and 3 days for shigellosis [ 8 ]. These diseases were selected because they were confirmed on the basis of laboratory criteria; they have the potential to occur in epidemics; they were designated nationally notifiable five years or more before the study period began; and the magnitude of reported disease incidence supported this analysis. Only finalized case-specific data reported from U.S. states and two autonomous reporting entities (New York City and Washington D.C., referred to as states, hereafter) that designated the reported condition as notifiable (reportable by law or regulation) and that met NNDSS publication criteria [ 9 ] were included in the analysis. Data were analyzed for MMWR years 1999, 2000, and 2001. Results Literature review Eight papers were identified that met the inclusion criteria for this study (Table 2 - see Additional file: 1 ) [ 10 - 17 ]. Seven of the eight papers met the inclusion criteria resulting from the literature review; an additional paper was identified from the review of reference lists of studies identified through the Medline search and studies citing CDC's evaluation guidelines [ 3 , 7 ]. Three of the eight papers in this study assessed national reporting timeliness; the remaining five papers focused on local or state reporting timeliness. The studies of national reporting timeliness focused on the following diseases: acquired immunodeficiency syndrome (AIDS) [ 17 ]; Neisseria meningitidis and Haemophilus influenzae infections [ 16 ]; and shigellosis, salmonellosis, hepatitis A, and bacterial meningitis [ 11 ]. The studies of local or state reporting timeliness analyzed data for AIDS [ 14 , 15 ], tuberculosis [ 13 ], influenza-like illness [ 10 ], and meningococcal disease [ 12 ]. In seven of the eight papers, timeliness was calculated as the median reporting delay between the date of disease occurrence (e.g., disease onset date, diagnosis date, or laboratory result date) and the date the public health system was notified or as the proportion of cases reported to the public health system in a specific time interval. In one study [ 10 ], epidemic curves were compared for two influenza surveillance systems and timeliness was assessed as the time interval between the epidemic peaks noted in each system. In addition, two studies described the factors associated with delayed reporting [ 13 , 15 ]. Seven of the eight studies addressed whether the calculated timeliness measure met the needs of the surveillance process being evaluated [ 10 , 12 - 17 ]. Measured timeliness was compared with recommended reporting timeliness in two papers – a national recommendation for local tuberculosis reporting timeliness [ 13 ] and a state mandate for reporting meningococcal disease cases to local public health [ 12 ]. The adequacy of the timeliness measure for the surveillance purpose was also assessed in other ways: 1) by comparing the timeliness of the same surveillance interval in an AIDS surveillance system before and after a major revision in the AIDS surveillance case definition [ 17 ], 2) by comparing the timeliness of the same surveillance interval across an active and a passive AIDS surveillance system [ 14 ], 3) by comparing outbreak detection abilities of an existing sentinel health care provider-based surveillance system for influenza-like illness with a new school-based system monitoring illness absenteeism [ 10 ], 4) by assessing whether reporting timeliness for Neisseria meningitidis and Haemophilus influenzae was adequate to initiate a rapid public health response [ 16 ], and 5) by comparing the timeliness of reporting by whether the case-patient's initial AIDS-defining condition was included in the 1997 or 1993 AIDS surveillance case definition [ 15 ]. The reporting timeliness of AIDS and bacterial meningitis (including meningococcal disease) surveillance systems were more frequently assessed than those for other infectious diseases. The AIDS reporting timeliness studies indicate that local and national AIDS reporting timeliness meets the goals of the AIDS surveillance systems monitoring trends, targeting prevention programs, estimating needs for medical and social services, and allocating resources [ 14 , 15 , 17 ]. Timeliness of AIDS surveillance improved after the revision of the AIDS surveillance case definition in 1993 [ 14 , 15 , 17 ]. Evaluation of Tennessee's Neisseria meningitidis infection surveillance system for 1989–1992 indicated that the lengthy reporting interval limited the usefulness of the system for supporting rapid response for control and prevention [ 16 ]. In contrast, a 1991 evaluation of New York State's meningococcal surveillance system indicated that the majority of cases (66%) were being reported within the recommended time frame (i.e., within one day of the diagnosis to ensure chemoprophylaxis for exposed persons) and therefore, supported prevention and control efforts [ 12 ]. In addition, on the basis of nationally notifiable infectious disease data from 1987, bacterial meningitis had the shortest reporting timeliness (median 20 days) of the other infectious diseases studied [ 11 ]. The definition of reference dates used in the timeliness evaluations varied. The initial date associated with the case varied among date of disease onset, date of diagnosis, and date of positive culture result. The ending date for the timeliness studies evaluated was the date the case report was received by the public health system, whether at the local, state, or national level. This time period corresponds to the sum of Intervals 1 and 2 or Interval 2 alone for local or state timeliness studies (Figure 1 ). For national evaluations of timeliness, the time period assessed was the sum of Intervals 1, 2, 3, and 4 or only Intervals 2, 3, and 4 (with or without inclusion of Intervals 5, 6, 7, and 8, dependent upon state protocol). NNDSS timeliness For MMWR years 1999–2001, a total of 9,276 cases of cryptosporidiosis, 12,332 cases of E. coli O157:H7 infection, 41,058 cases of hepatitis A virus acute infection, 7,090 cases of meningococcal disease, 22,735 cases of pertussis, 120,688 cases of salmonellosis, and 60,693 cases of shigellosis and were reported to NNDSS. Of those, 7,079 (76.3%) cryptosporidiosis case reports, 9,674 (78.4%) case reports of E. coli O157:H7 infection, 32,953 (80.3%) case reports of acute hepatitis A virus infection, 5,580 (78.7%) case reports of meningococcal disease, 19,904 (87.5%) case reports of pertussis, 84,746 (70.2%) case reports of salmonellosis, and 41,643 (68.6%) case reports of shigellosis were eligible for analysis. A total of 72,293 (26.4%) case reports were excluded for one or more of the following reasons: reported as a summary or aggregate record in which individual cases may have different event dates (20,194 cases), unknown or missing date types (20,019 cases), date type coded to MMWR report date (11,851 cases), and calculated reporting lag had a value of zero (indicating the event date and midpoint of the MMWR week matched) or had a negative value (indicating the event date was later than the mid-point of the MMWR week [67,557 cases]). Timeliness of reporting varied by disease and date type (Table 3 ). For cases reported with a disease onset date, the median reporting delay across all reporting states varied from 12 days for meningococcal disease to 40 days for pertussis. For cases reported with a laboratory result date, median reporting delay varied from 10 days for both meningococcal disease and shigellosis to 19 days for pertussis. There was also substantial variation in state-specific median reporting delays for each disease (Table 3 ). For example, for meningococcal disease cases reported with a laboratory result date, state-specific median reporting delay varied from a median of 2 days in one state to 117 days in another. Table 3 Timeliness of reporting of selected nationally notifiable infectious diseases, by date type, NNDSS, 1999–2001 Date type (Intervals from Figure 1) Disease (incubation period), Characteristic Disease onset (Intervals 1,2,3,4) Diagnosis date (Intervals #2,3,4) Lab result date (Intervals #2,3,4) Date of first report to the community health system (Intervals #3,4) Cryptosporidiosis (7 day incubation period) Median time interval (days) 22 14 13 26 State-specific reporting range a 2–149 1–73 2–58 1–53 No. cases 4,130 956 1,825 168 No. states 44 24 41 15 % within 1, 2 incubation periods b 24%, 39% 37%, 50% 35%, 54% 19%, 33% E. Coli O157:H7 (4 day incubation period) Median time interval (days) 17 21 11 15 State-specific reporting range a 2–81 2–41 1–53 1–49 No. cases 6,891 473 2,206 104 No. states 48 22 39 14 % within 1, 2 incubation periods b 15%, 27% 13%, 25% 19%, 39% 21%, 33% Hepatitis A, acute (30 day incubation period) Median time interval (days) 23 18 12 12 State-specific reporting range a 2–54 2–80 2–29,231 + 1–126 No. cases 21,570 4,394 6,695 294 No. states 49 36 39 14 % within 1, 2 incubation periods b 62%, 84% 67%, 83% 82%, 94% 79%, 91% Meningococcal disease (4 day incubation period) Median time interval (days) 12 13 10 10 State-specific reporting range a 2–56 1–54 2–117 4–62 No. cases 3,804 450 1,255 71 No. states 50 30 39 7 % within 1, 2 incubation periods b 23%, 39% 26%, 40% 25%, 44% 31%, 42% Pertussis (20 day incubation period) Median time interval (days) 40 31 19 23 State-specific reporting range a 2–124 1–106 2–190 2–48 No. cases 18,750 289 758 107 No. states 50 26 34 15 % within 1, 2 incubation periods b 24%, 50% 34%, 60% 53%, 78% 45%, 68% Salmonellosis (1.5 day incubation period) Median time interval (days) 17 7 12 16 State-specific reporting range a 2–44 1–54 2–61 1–27 No. cases 49,659 5,558 28,172 1,357 No. states 47 35 42 28 % within 1, 2 incubation periods b 4%, 13% 17%, 43% 6%, 17% 7%, 19% Shigellosis (3 day incubation period) Median time interval (days) 15 10 10 9 State-specific reporting range a 2–43 1–51 2–34 1–26 No. cases 26,635 2,850 11,603 555 No. states 46 28 41 17 % within 1, 2 incubation periods b 15%, 22% 33%, 39% 22%, 35% 29%, 41% *Source: Control of Communicable Diseases Manual 17 th Edition [8]. + The maximum state-specific median reporting delay for this disease and date type is from a state that reported 19 cases having event years 1919 or 1920. Excluding these cases as data entry errors, the maximum state-specific median reporting delay is 78 days. a State-specific median reporting range (minimum, maximum) in days b % of cases reported within 1 and 2 incubation periods, respectively For the same date type, NNDSS diseases with longest incubation periods tended to have a higher percentage of cases reported within one or two incubation periods than NNDSS diseases with shorter incubation periods (Table 3 ). For example, for acute hepatitis A virus infection, which had the longest incubation period of all the study diseases, more than 60% of cases were reported within one incubation period, for each date type reported. For all other diseases except pertussis, less than 40% of cases were reported within one incubation period for each reported date type. For pertussis, the percentage of cases reported within one incubation period varied from 24% for reports with disease onset date to 53% for case reports with laboratory result dates. In addition, state-specific percentage of cases reported within one or two incubation periods varied for a given disease and date type (data not shown). Comparison of NNDSS timeliness and literature review results The 1999–2001 NNDSS meningococcal disease median reporting interval between date of disease onset and date of report to CDC in this study was 8 days shorter than a previous study reported [ 11 ] using 1987 notifiable disease data for bacterial meningitis (median 20 days); and, the meningococcal disease median reporting delay was 9 days shorter in this study than in a previous study [ 16 ] using Tennessee's data for the years 1989–1992 for Neisseria meningitidis infection (median 21 days). In addition, the median reporting delay between disease onset and the date of report to CDC was shorter in this study than in a previous study (which used 1987 notifiable disease data) by 10 days for hepatitis A, 5 days for salmonellosis, and 8 days for shigellosis [ 11 ]. Discussion Few published studies evaluating surveillance systems presented timeliness measures. When timeliness was evaluated, standard methods were not used. Information collected by public health surveillance systems should support the quantitative assessment of timeliness by various steps in the pubic health surveillance process. Public health programs should periodically assess timeliness of specific steps in the surveillance system process to ensure that the objectives of the surveillance system are being met. A more structured approach to describing timeliness studies should be considered. Published papers describing local or state surveillance system reporting timeliness generally do not explicitly describe the surveillance system processes contributing to the timeliness measure, such as processing and analyzing the data or implementing a public health action before data are reported from a state to CDC. To facilitate future comparisons of reporting timeliness across jurisdictions, studies should include an explicit description of the public health surveillance reporting process and the surveillance process interval being measured. Additionally, surveillance information systems must support the collection of appropriate reference dates to allow the assessment of the timeliness of specific surveillance processes. A more structured approach to describing timeliness studies could include a description of the following characteristics: 1) the level of the public health system being assessed (e.g., local, state, or national), 2) the purpose of the surveillance evaluation, 3) goals of the surveillance system, 4) the surveillance interval being measured and a description of the reference dates that define the upper and lower boundaries of the surveillance interval, 5) the surveillance steps (processes or activities) that contribute to the surveillance interval being measured, 6) whether the measured timeliness met the needs of the surveillance step being evaluated, and 7) whether the timeliness met the goals of the surveillance system. No single timeliness measure will achieve the purpose of all evaluations or meet all the goals of the surveillance system. In addition, if the goal of the surveillance evaluation is to identify ways to improve timeliness, the analysis should identify factors associated with delayed reporting, such as the role of specific case ascertainment sources. The 1999–2001 national notifiable diseases data were timely enough to support the following surveillance objectives: monitoring trends over time, informing allocation of public health resources, monitoring the effectiveness of disease control, identifying high risk populations, and testing hypotheses. If NNDSS data are to be used to support timely identification of and response to multistate outbreaks at the national level, the timeliness of reporting needs to be enhanced for all diseases, but especially for diseases with the shortest incubation periods (e.g., cryptosporidiosis, E. coli O157:H7, meningococcal disease, salmonellosis, and shigellosis). Until reporting timeliness is enhanced, the application of aberration detection analytic methods to NNDSS data to aid in the identification of changes in disease reporting that may indicate a multistate outbreak in time to alert states for the purposes of disease control and prevention may be of limited use. Future work to improve reporting timeliness will need to address the substantial variation across states. As states enhance their reporting mechanisms with the use of automated electronic laboratory reporting systems [ 18 ], there may be less variation in state-specific reporting timeliness, but this should be assessed. NNDSS timeliness improved compared to timeliness of notifiable infectious diseases measured in previous reports [ 11 , 16 ]. However, the methods or variables used in these analyses were different. A few factors may have contributed to improvements in timeliness seen in this study. Since 1992, states have been routinely transmitting electronic case-specific records intended to improve reporting procedures and protocols. In addition, the use of automated electronic laboratory reporting to enhance infectious disease case reporting may have contributed to increased timeliness. Our study findings are subject to several limitations. The variables available for assessing NNDSS reporting timeliness are based on the MMWR week numbers that are assigned by states and the earliest known date reported in association with the case. While these variables might provide an estimate of national reporting timeliness, NNDSS data do not include a fixed date defining when a case report was initially transmitted to CDC or received at CDC, which would provide a more precise measure of national reporting timeliness. NNDSS data management protocols should be modified to permit direct calculation of national reporting timeliness. If the ability to support outbreak detection at the national level using NNDSS data is generally viewed as an important and sustainable enhancement for the NNDSS, states and CDC programs should facilitate reporting that more closely approximates real-time and define reporting protocols and data requirements to ensure that reporting timeliness can be improved and accurately monitored. The current NNDSS practice of weekly reporting and data processing limits reporting timeliness to CDC. Lastly, 72,293 (26.4%) cases were excluded from our analysis because the information contained in the database would not permit calculation of timeliness and this exclusion may have resulted in our study results either falsely overestimating or underestimating the magnitude of NNDSS reporting lags. The reporting timeliness variations across states may result from different reporting protocols in the states (e.g., centralized versus distributed reporting within the state's public health system) or from variations in how states assign MMWR week numbers. Other factors that might have contributed to reporting delay in our study included: the patient's recognition of symptoms; the patient's acquisition of medical care; the use of confirmatory laboratory testing; reporting by the health care provider or the laboratory to the local, county, or state public health authority; the volume of cases identified in the state; case follow-up investigations to verify the case report or to collect additional case information; periods of decreased surveillance system activity due to variable staffing levels; computer system down-time for maintenance, upgrades, or new application development; and data processing routines, such as data validation or error checking. Following a structured approach to evaluation of timeliness by specifying the surveillance objectives and the process(es) being measured may allow better definition of the factors that contribute to reporting delay. It was beyond the scope of this study to assess how these factors contribute to NNDSS reporting timeliness. In addition to reporting timeliness, other surveillance system attributes are important to assess (e.g., completeness of reporting). Completeness of notifiable infectious diseases reporting in the United States varies from 9% to 99% [ 7 ]. Six of the eight papers reviewed for this study assessed completeness of reporting [ 12 - 17 ]. One paper [ 14 ] noted that although the timeliness of the AIDS passive and active surveillance systems were comparable, the completeness of the active AIDS reporting system far exceeded the reporting completeness for the passive system. This highlights the importance of evaluating completeness and timeliness and other surveillance system attributes concurrently, before contemplating any changes to a surveillance system based on the assessment of a single attribute. To improve public health surveillance infrastructure and performance in the United States, CDC and local and state health agencies are integrating a number of public health surveillance systems monitoring infectious diseases in the United States, including the NNDSS, into the National Electronic Disease Surveillance System (NEDSS) [ 19 , 20 ]. NEDSS outlines a standards-based approach to disease surveillance and intends to connect public health surveillance to the clinical information systems infrastructure. As a result, NEDSS promises to improve the accuracy, completeness, and timeliness of disease reporting to state and local health departments and CDC. Conclusions To facilitate comparisons of surveillance system timeliness studies across jurisdictions or health conditions, a more standardized approach to describing timeliness studies is warranted. Public health surveillance systems should ensure that timeliness can be measured for specific surveillance system processes and in the context of the goals of surveillance. In addition, when timeliness is being measured, it is important to be explicit about how it is being measured. Our analysis of NNDSS reporting timeliness suggests that current acute hepatitis A infection reporting timeliness may be sufficient to support a timely public health response in the event of a multistate outbreak. However, for the other conditions evaluated, the long reporting lag and the variability across states limits the usefulness of NNDSS data and aberration detection analysis of those data for identification of and response to multistate outbreaks. The NNDSS timeliness data presented in this paper represents a baseline against which timeliness can be measured in the future. Further study is needed to identify the major sources of reporting delay and to assess how NNDSS reporting timeliness may be improved for the timely detection of cases and disease clusters. Competing interests None declared. Author's contributions Both authors contributed equally to project conception and write-up of the manuscript. RAJ was responsible for data analysis. Both authors read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here: Supplementary Material Additional File 1 Published reports quantitatively measuring timeliness of reporting infectious disease surveillance data. Table summarizes the findings of the review of published literature about quantitative measurements of infectious disease surveillance system timeliness Click here for file	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC509250.xml
517708	SED, a normalization free method for DNA microarray data analysis	Background Analysis of DNA microarray data usually begins with a normalization step where intensities of different arrays are adjusted to the same scale so that the intensity levels from different arrays can be compared with one other. Both simple total array intensity-based as well as more complex "local intensity level" dependent normalization methods have been developed, some of which are widely used. Much less developed methods for microarray data analysis include those that bypass the normalization step and therefore yield results that are not confounded by potential normalization errors. Results Instead of focusing on the raw intensity levels, we developed a new method for microarray data analysis that maps each gene's expression intensity level to a high dimensional space of SEDs (Signs of Expression Difference), the signs of the expression intensity difference between a given gene and every other gene on the array. Since SED are unchanged under any monotonic transformation of intensity levels, the SED based method is normalization free. When tested on a multi-class tumor classification problem, simple Naive Bayes and Nearest Neighbor methods using the SED approach gave results comparable with normalized intensity-based algorithms. Furthermore, a high percentage of classifiers based on a single gene's SED gave good classification results, suggesting that SED does capture essential information from the intensity levels. Conclusion The results of testing this new method on multi-class tumor classification problems suggests that the SED-based, normalization-free method of microarray data analysis is feasible and promising.	Background DNA microarray technology is now playing an increasingly important role in biomedical research. Microarray technology gives one the opportunity to measure gene expression levels of thousands to tens of thousands of genes simultaneously, in order to study the differential gene expression pattern between different developmental stages, diseases states and samples treated with drugs or other compounds. Before comparing data from different arrays to address these biological questions, however, a "much more mundane but indispensable " normalization step [ 1 ] is currently used in most microarray analyses. Because of the slight difference in RNA quantities, imaging settings and other variables, even in very controlled experiments the intensity levels from different arrays are of different scales and need to be normalized before they can be compared with each other. Various normalization methods have been developed and some are widely used. The simplest method is total intensity based normalization [ 2 ]; this approach scales intensity levels of every gene by a constant factor so that total intensities of all the arrays are the same. "Spiked-in" based normalization methods scale intensity based on spiked-in standards [ 3 ]. Nonlinear normalization methods use local regression to scale intensities to compensate for the intensity-dependent differences between arrays [ 4 - 6 ]. For most current applications, these normalization methods seem to be adequate. However, the residual left by a less than perfect normalization procedure is another source of non-biological variation that is usually non-desirable, especially when the differences in expression levels are expected to be small [ 7 ]. In addition, if the goal is meta-analysis of multiple sets of microarray data [ 8 , 9 ] systematic differences between experiments may result in a normalization artifact. We were therefore interested in developing an approach to analyse microarray data without first performing a normalization step. Our approach was partly inspired by non-parametric statistical methods [ 10 ]. For example, nonparametric methods that use ranks [ 11 , 12 ] to compare microarray results, in addition to being distribution free, have the additional advantage of being normalization free. DNA microarray technology has been used widely in biomedical studies. One interesting application is in the area of molecular classification; one popular use is in the comparison of tumor samples. Since clinical and histopathological classification is sometimes difficult and labor-intensive, the use of genome wide expression patterns to classify tumor samples has recently become a very active research area [ 13 - 16 ]. Although some tumors appear to be amenable to classification using microarray data [ 17 , 18 ], general multiple tumor classification using microarray data has proved to be an interesting and challenging task for several reasons: the general difficulties inherent in multi-class classification problems, the small number of samples available, and the inherent biological variation between specimens, etc . We decided to use multi-class tumor classification as a test case to illustrate the power of our approach. We compared our results for a multi-class tumor classification problem with more conventional approaches published by Ramaswamy et al . [ 19 ] and Yeang CH et al. [ 20 ]. These authors compared the accuracies of using k-Nearest Neighbors (kNN, 60–70%), Weighted Voting (WV, 60–70%) and Support Vector Machine (SVM, 80%) algorithms in a multi-class tumor classification problem and concluded that SVM is a more powerful machine learning algorithm for this application. Results Normalization Free approach to microarray data analysis Generally, measurements on single microarrays give a real-valued intensity level x i (1<= i <= N) for each gene i on the array, where N is the total number of genes on the array. Without first doing some type of normalization, the intensity level of gene i from array A, x i A , cannot be directly compared with the intensity level of gene i from array B, x i B . In this study, we sought an alternative quantity or quantities that can be directly compared between different arrays without compromising important biological information. One obvious candidate is r i , the rank of intensity level of gene i on the array. However, we felt that rank is not an adequate measure because information about relative expression level is not represented explicitly. Instead, we decided to use the following measures. Let s ij = 1 if(x i - x j > 0) 0 otherwise (1) , where 1<= i, j <= N. Basically, s ij is the sign of the intensity difference of gene i and j on a single microarray and therefore will remain unchanged under any monotonic transformation of x. Therefore, instead of computing with the absolute expression level of a gene, its relative level to all the other genes on the microarray is used. For each gene i, instead of one real valued x i , the approach uses s i = (s i1 , ..., s ij , ..., s iN ), a binary vector of size N. For ease of reference, we will simply refer to this value as the SED (Signs of Expression Difference) of gene i; and the entire matrix (s ij ) the SED of the array. Given (x i ), s ij is simply and uniquely defined but (s ij ) does not uniquely determine x i so some information is lost by only using (s ij ) instead of (x i ). Since r i = Σ j = 1,...,N s ij is the rank of gene i in terms of intensity levels, rank information is preserved in (s ij ). What is lost in the transformation from (x i ) -> (s ij ) is just the intensity differences between the closest ranked genes, which in most cases are small, considering that microarray data are generally considered "very noisy". It was our major goal to demonstrate that (s ij ) has indeed captured important components of the information from (x i ). Instead of directly using the intensity levels x, and its derivatives such as the mean μ, the standard deviation σ and the signal to noise ratio (S2N) between two sample groups A and B, [(μ A + μ B )/(σ A + σ B )], we will use (s ij ) to compare gene expression differences between arrays. Since we expect measurement variations within an array will be less than those between arrays and we take the signs of relative expression differences to get SED, we expect the SED will be less "noisy". However, the value of any single s ij may still vary between technical and biological replicates. One would expect more s ij would change values randomly if the technical replicates was done on arrays that were fabricated in a different run than arrays from same run, for example. Biological variations are expected to be even more frequent. However, we hypothesize that we can perform statistical analysis on the SED, which contains tens of thousands of s ij for a single gene, and minimize the impact of such noises. We will also consider (sp ij ), a natural generalization of the SED concept. Here, sp ij is the probability of x i > x j . In other words, imagining one can get a large number, n, of either technical or biological replicates of the sample of interest, then sp ij = m/n as n -> 8, where m = Σ K = 1,...,n s ij K and s K is for replicate K. We will call (sp i ) = (sp i1 , ..., sp ij , ..., sp iN ) the SED probabilities of gene i. Note that in calculating both SED and SED probabilities, only intensity comparisons within arrays are involved and therefore forego the normalization step. For example, if gene i is more highly expressed in sample A than B we would expect that more s ij A than s ij B would be 1 instead of 0 and the overall (very loosely defined) sp ij A would be larger than sp ij B . Since rank can be calculated from SED, any rank based method can be expanded to use SED. A gene i's SED can be viewed in two different perspectives. On one hand, it provides information about gene i's expression level relative to every other gene on the array, and therefore can be used to examine gene i's expression patterns between samples. On the other hand, it also provides information about the expression levels of all the other genes on the array, using the gene i as a control, in essence. Therefore, SED can be used to study questions either at the gene level or at the array level. In this paper, we focus on solving a simpler problem at the array level where it is not necessary to decide whether the expression level of an individual gene is increased or decreased between array A and B and by how much. Rather, it is focused on whether the overall expression patterns are different at all between array A and B. Multi-class classification of tumor samples To test whether (s ij ) and (sp ij ) extracts most of the information from (x i ), we used these values in a test case of a multi class classification problem described by Ramaswamy et al. and Yeang et al [ 19 , 20 ]. Two algorithms were used to classify each of the 144 tumor samples into one of 14 tumor classes. One is the Naive Bayes (NB) classifier [ 21 ] using SED probabilities. The other is the Nearest Neighbor (NN) classifier using SED. In the NB method, to classify a sample T, we first calculate (sp ij C ) for each class C (1 <= C <= 14) using training samples, i.e. the 144 samples with T taken away (for details see methods). Then sample T is classified according to: score(T, C) = Σ i = 1,...N Σ j = 1,...,N log(p ij ), (2) where p ij = sp ij C if s ij T = 1 1 - sp ij C otherwise T is simply classified to the class C that has the maximum score. In the NN method, we compute instead, for each training sample t, matches(T, t) = Σ i = 1,...N Σ j = 1,...,N δ(s ij T , s ij t ), (3) where δ(x, y) = 1 if x = y. 0 otherwise Then T is classified to class C of the sample t that has the maximum matches. If one is to give a statistical interpretation of these scores, one can simply view (sp ij ) as defining a multi-binomial probability model. In addition, one could consider each s ij as a draw from a binomial distribution with probability p ij = sp ij . Then, the score(T, C) is simply a logarithm of the probability p to get all the s ij exactly the same as s ij T under the probability model C where p ij = sp ij C . (Since each class defines a different probability model, score(T,C) for difference class C, in theory, should not be directly compared. Instead, a P value should be calculated from the probability model for Pr(p<exp(Score(T, C))) and used to evaluate the closeness of sample T to each class C. For simplicity, we are not considering such issues here.) When the NB algorithm was applied to the 144 samples, the accuracy obtained was about 63%; the NN algorithm performed slightly better and gave an accuracy of 70%. Feature selection Depending on the algorithm, a better classification result can sometimes be obtained by using a subset of genes [ 22 , 23 ]. We were interested to know whether feature selection helps to increase accuracy in our approach. Within our framework, it is easier to treat (i,j) pairs as selection units. We therefore filtered out (i,j) pairs where the variance of sp ij across the 14 tumor classes was less than a pre-determined value and left the rest of the algorithm unchanged. We reasoned that the filtered-out part of the matrix has less discriminating power across the tumor spectrum and might add noise due to the small sample sizes used. Using the NB algorithm, the best result achieved was 70% with a cutoff σ 2 = 0.06, while the NN method with σ 2 = 0.05 gave 77%. Table 1 lists the accuracies achieved using different feature cutoffs. "Single gene's SED" based classifier The above procedure utilized the entire SED matrix as a classifier. In other words, all the relations between genes were considered in the classification. To determine whether the inclusion of the whole matrix was actually required to achieve the current accuracy, we investigated the efficacy of using single gene's SED as classifiers. In these cases, we define a classifier based on gene i and its relative expression to every other gene. Therefore, the score i (T, C) = Σ j = 1,...,N log(p ij ), where p ij is as same as mentioned previously. Similarly, matches i (T, t) = Σ j = 1,...,N δ(s ij T , s ij t ) defines a classifier based on gene i's SED. Fig. 1 shows a display of the cumulative frequency of single gene SED based classifiers versus accuracy for all the 16063 classifiers. In general, single gene-based classifiers performed worse than the whole genome-based classifiers, as expected. Nevertheless, most of the classifiers performed reasonably well, compared with just using single gene expression levels. About 80% of the single gene based classifiers resulted in an accuracy between 40% and 60%, while about half of the classifiers had an accuracy greater than 50%. These results suggest that there is a lot of redundant information in the SEDs and SED probabilities and that our method should be reasonably robust. We then investigated the number of genes that are required to achieve the current accuracy. Fig. 2 shows the combined results for classifiers using only a subset of genes. Our results suggest that a subset of genes (~200) is sufficient for predictions and that the prediction accuracy is stable after 1000 genes. Different classification accuracy between tumor classes From the analyses described above, we noticed that there was a significant difference in accuracy between different tumor classes. For 3 classes (LY, LE, CNS) we obtained either 100% or close to 100% accuracy (see Table 2 for detail). Since these happen to correspond to the 3 classes with more than double the number of training samples than the other classes, we tested whether this high accuracy is due to the larger sample sizes by using only 8 training samples for every tumor classes. The results were essentially the same, indicating that sample size is not the issue. On the other hand, there are classes where we obtained very poor results; these often happen to be the same classes where SVM in [ 19 , 20 ] performed poorly as well (see Table 2 ). We were interested in exploring the possible reasons for misclassification. In Fig. 3 , a scatterplot of the SED Match scores (without feature selection) between 8 OV samples and 8 CNS samples is displayed. The OV and CNS class were selected since one is "very hard" and the other is "very easy" to classify. Without trying to be statistically correct, the plot does suggest that samples of the OV class are in general "farther away" from each other, compared with those from the CNS class. As this may be one of the reasons that the OV class is harder to classify, algorithms that take this kind of information into account may perform better than the simple ones we have presented here. Discussion Although we used the multi-class tumor classification problem as our test case, our major goal was to illustrate the feasibility of the normalization free SED approach, and not in sample classification per se . Therefore, we chose the algorithms NB and NN for their simplicity and not for their performance in solving this specific problem. The performance of a classifier depends, in this case, mainly on the power of its algorithm, and the data representation it used. From a machine learning perspective, one can simply view the intensity -> SED transformation as a change in data representation, a mapping from the gene's attribute, intensity x, to some features SED. It was our goal to demonstrate that the new features (SED and SED probability), in addition to being normalization free, still convey the essential information in the original attribute, the intensity x. Since the data representation is quite different between the intensity x (a real valued quantity) and SED (a binary valued vector of rather large size N), it is difficult to directly compare the two. No obvious yet non-trivial algorithms work with both representations; even if there were such an algorithm it is not clear that it would be the right one to use for comparison as it might well be the case that different data representation works best with different algorithms. Here, we have limited the presentation to some empirical results with SED representation, which are comparable with results using several different algorithms that are based directly on raw intensity [ 19 , 20 ]. Our classification results are close to those obtained with WV and kNN methods, which are based on directly focusing on intensity levels. Previous results using SVM were significantly better, but we feel the differences are due more to the power of the algorithm [ 24 ] than the way information is coded. In fact, slightly more accurate results are obtained with modification of algorithms that directly manipulate intensity levels [ 25 , 26 ]. We do not imply that the algorithms (NB and NN) we chose are better than other alternatives (and we do not have empirical evidence pointing either way). Instead, we fully expect more sophisticated algorithms would work better with the SED approach as well. Certainly, SED probability is more information rich than SED. We expect that an SED probability based analysis would perform better than the simple binary valued SED. In this paper, we mainly tested the SED. SED probability is only used for a group of samples, not for single samples. If one limits oneself to use only raw data, then for single arrays one can only get SED. However, if some assumptions about the patterns of gene expression levels can be made, one can certainly get an estimation of SED probability even for a single array. For example, as in some nonlinear normalization algorithms, if one assumes that the variation of expression levels are similar for genes with similar expression levels, then one can estimate SED probability from a probability model. Also, the magnitude of the intensity difference can also be used to help such an estimation. Alternatively, as more and more microarray data become available, one can use other similar samples to get an estimation of a prior SED probability, and then use a Bayesian approach to estimate the sample's SED probability. The obvious disadvantage of our SED based approach is that for each gene expression level, one is not dealing with a single real number but instead a vector of size N, where N is in the tens of thousands. This could significantly increase both computing time and memory requirement (however, see methods for details) On the other hand, it also has certain advantages: 1) It is free of normalization noise. Since it is generally believed that biological variation is larger than technical variation and normalization noise is just another source of technical variation, the benefit here is only of a limited scope. However, it may be important when the expression level difference one is interested in is small. 2) In addition to being normalization free, SED and SED probability also have the advantage in being distribution free, and therefore could perform better if the intensity levels were non-normal. 3) SED and SED probability are easier to interpret. SED values can easily be checked against raw intensity levels according to Eq. (1). While SED probability is one step further away from intensity levels, one could still have an intuitive sense of it and make comparisons between different experiments. It would be much harder to have a real grasp of the absolute gene expression level, except that it is "high" or "low" or somewhere "in between"; it is certainly harder to compare between experiments intuitively. We have only tested the SED approach on datasets that are from the same chip format. Data from different chip formats or complete different technology platforms, of course, would be harder to compare. But they are also challenge for normalization based method. It would certainly be interesting to compare SED and normalization based method under these more challenging conditions. If this normalization-free approach (SED) proves to retain the essential biological information in general, its application may be extended to meta-analysis where different datasets could be integrated and intervalidated. The method could also be used when the number of arrays is a limiting factor for experiments. For example, one could take advantage of the massive amount of public array data, obtain prior distribution of SED probabilities from datasets with similar conditions, and analyze new data within a Bayesian framework. If the performance of the nearest neighbor method in general is anywhere close to what we demonstrated in the multi tumor classifications here (as is clear from Eq. (3) and Fig. 3 , the nearest neighbor method, without feature selection at least, allows direct sample vs. sample comparison. Note also that the samples in the multi tumor problems are from different biological specimens, therefore, large between-sample-variation is expected), it might be used as a microarray database query method, i.e. , to find similar microarray results in the database that are "similar" to one's own, independent of array annotations. It might also be worth noting that the SED approach could easily be applied to other kinds of comparative data analysis for samples with very large numbers of "noisy" attributes. The SED approach may also perform better when between-sample-variation is large, especially if such variation contains some rather uninteresting technical measurement errors that would not affect within-sample-variations. Conclusions We have proposed a new approach to analyze microarray data and tested the method on a set of publicly available datasets. The results were comparable to those obtained with some widely used normalization based algorithms. We hope that we have demonstrated that this normalization free method is feasible and promising. We think the SED based, normalization free approach could be used to complement the more popular normalization based approaches in microarray data analysis. Methods Microarray data for multiple tumor samples were downloaded from . Naive Bayes and Nearest Neighbour Classifiers were implemented in the Java programming language. Ad hoc analysis was done with perl scripts. Graphics were generated using the R computing environment. Naive Bayes method Because of the uneven and relatively small sample sizes for each tumor class (mostly 8 but up to 24), extra care was taken in computing sp ij . Assuming a prior probability of 0.5, sp ij was estimated by Bayesian posterior probability (m+1)/(n+2) where n is the total number of samples in the class and m is the total number of samples where x i > x j . For classes that were over-represented (sample size > 8), the threshold of sp ij was set to [0.125, 0.875], since the NB method is sensitive to the extreme values of sp, and samples can be over-predicated without thresholds. In addition, several alternatives were tested to demonstrate that our results were reasonably robust and not sensitive to the particular choices we made: 1) To examine the influence of the sample size, in a separate analysis the sample size of ME, LE, CNS class was artificially reduced to 8, i.e. only the first 8 samples were used to calculate sp ij with no significant change of results observed; 2) Since sp ij depends on the sample size n for each tumor class, we have applied a "sample replacement" strategy in addition to the usual "take-one-sample-out" approach for cross-validation, i.e. when one sample is taken out as the test sample, another sample from the same class is duplicated to take its place to keep the sample size constant. Essentially the same results were obtained. Results reported are from the sample replacement runs. In Feature Filtering, the variance of sp ij between all 14 classes was calculated as: σ 2 = (Σ C = 1,...,14, sp ij C * sp ij C )/14 - ((Σ C = 1,...,14 sp ij C )/14) 2 (4) with the test sample taken out, and used as the criterion for feature exclusion. Nearest neighbor method Feature filter was done as in the Naive Bayes Method. Software implementation and availability The analysis was done on a computer (Pentium M 1.5 GHz) operating under Microsoft XP. Both Naïve Bayes and Nearest Neighbor Classifiers are implemented in Java. Since SED can be easily calculated from the raw intensities only the later are kept in memory and SED are computed from the intensities on an as-needed bases. Memory needed to analysis the 144 samples is less than 64 MB. The most computationally intensive algorithm that we tried is the Nearest Neighbor method without any feature selections and it takes about 10 sec to calculate SED score for one pair of tumor samples with about 16000 genes. A Java program named SED (including source code) to perform nearest neighbor analysis of microarray samples is freely available by contacting author at hw14@columbia.edu . Authors' contributions H.W. conceived of the SED study and performed implementations. H.H. refined the approach and provided additional statistical insight on SED. Both authors read and approved the final manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC517708.xml
514573	"Is there nothing more practical than a good theory?": Why innovations and advances in health behavior change will arise if interventions are used to test and refine theory	Theoretical and practical innovations are needed if we are to advance efforts to persuade and enable people to make healthy changes in their behavior. In this paper, I propose that progress in our understanding of and ability to promote health behavior change depends upon greater interdependence in the research activities undertaken by basic and applied behavioral scientists. In particular, both theorists and interventionists need to treat a theory as a dynamic entity whose form and value rests upon it being rigorously applied, tested and refined in both the laboratory and the field. To this end, greater advantage needs to be taken of the opportunities that interventions afford for theory-testing and, moreover, the data generated by these activities need to stimulate and inform efforts to revise, refine, or reject theoretical principles.	Background Even with the dramatic advances in our understanding of the biological processes that determine health and illness, it has never been more clear that rates of disease morbidity and premature mortality reflect people's behavioral practices. [ 1 ] The benefits, both for individuals and the societies in which they live, that would come from systematic improvements in diet, physical activity, and use of substances such as tobacco, alcohol, and illicit drugs are tantalizing and provide ample motivation to develop initiatives to elicit changes in health behavior. Yet, health behavior change has proven a worthy adversary. Despite the commitment of considerable time and effort, innovations and advances in our ability to improve health behaviors have been modest. In particular, the specification of methods that produce sustained improvements in behavior have been elusive [ 2 - 5 ]. At the same time, innovations in theories of health behavior have also been modest. Investigators continue to advocate for a broad range of theories and there has been limited progress in demonstrating the unique value of any specific theory. [ 6 - 8 ] Although there may be consensus in the professional community that there are considerable gaps in our understanding of health behavior change, critiques of the current state of affairs more often that not reflect the professional interests of the critic. Investigators who strive to specify the structural and psychological processes that regulate people's behavior lament the fact that too many interventions are not guided by a theoretical framework that specifies how they are supposed to elicit health behavior change. At the same time, investigators who design and implement health behavior interventions lament that the preponderance of theories of health behavior make it difficult to discern what factors are likely to be the most effective targets for intervention. Moreover, it is argued that theories are not sufficiently specified to determine when or how to modify factors that are to be targeted in an intervention. Of course, concerns regarding the link between theory and practice are not new and efforts to address this problem have taken several forms. Considerable effort has been given to provide practitioners with a comprehensive and concise understanding of the array of theories that have been developed to address health behavior. [ 9 ] Moreover, conceptual frameworks such as PRECEDE-PROCEED [ 10 ] and Intervention Mapping [ 11 ] have been developed to provide investigators with a structured process to improve the accuracy and ease with which theoretical concepts are used to address a practical problem. In both cases, these efforts have targeted improving how theoretical principles are applied and, in doing so, have relied on the assumption that current theories of health behavior are useful and productive. Is this assumption valid? Could the often repeated plea for investigators to ground their intervention efforts in theory be a sign that there are significant limitations to the practical principles that can be derived from current theories of health behavior? If so, merely improving how people use theories will not be sufficient. What is needed is a shift in how we engage the interplay between theory and practice, with an emphasis placed on developing initiatives that target opportunities to develop, test, refine health behavior theory. In this paper, I describe and advocate for a model of collaboration between basic and applied behavioral scientists. Although I recognize the value of improving the manner in which theoretical principles are matched to problems and methods, I propose that innovations in our understanding of and ability to promote health behavior change will not arise if theory is construed as a fixed entity that is delivered to interventionists for implementation. To date, although theories may fluctuate in their popularity, their properties have remained strikingly static over time. I believe greater attention must be paid to refining and, when necessary, rejecting theoretical principles. For this process to take shape, there needs to be an on-going series of exchanges between theorists and interventionists in which theory is treated as a dynamic entity whose value depends on it being not only applied and tested rigorously, but also refined based on the findings afforded by those tests. A fundamental implication of this perspective is that improvements in both health behavior theory and intervention methods depend on each other. If investigators are more receptive to the opportunities interventions afford for theory testing, there will be a dramatic increase in data that can reveal the adequacies and inadequacies of a given theory. These data will, in turn, enable theorists to improve the quality of the theoretical models available to guide subsequent intervention efforts. Discussion When is an intervention effective? Interventions are designed to address important practical problems (e.g., obesity) and thus their value is inextricably linked to their ability to alleviate the targeted problem. Interventions need to provide a meaningful return on the time, money, and effort invested such that the outcomes afforded by a intervention strategy are proportional to the resources utilized. Of course, determining what is a sufficient return on an investment can be a challenge. Small effects may be impressive if the intervention is directed at a construct or behavior that is considered difficult to move. [ 12 ] In addition, interventions can have minimal impact on an individual's behavior but when disseminated widely have a dramatic impact at the societal level. [ 13 ] What conditions are likely to facilitate a successful intervention? Broadly speaking, an intervention is most likely to be effective if it is appropriately grounded in the practical problem targeted. [ 11 ] For example, consider an intervention to promote healthy food choices. The intervention design team must possess a clear understanding of who is engaging in the targeted behavior (e.g., who is making unhealthy food choices), the underlying nature of the behavior (e.g., the frequency and function of food choices), and the context in which the behavior is performed (e.g., where and with whom do people make choices about food). In a similar manner, the intervention needs to be appropriately grounded in the biological, structural and psychological processes that shape and regulate people's behavioral practices. [ 14 - 16 ] For example, the expected value of altering a feature of the environment in which people make food choices (e.g., increasing the cost of high-fat foods) is predicated on the assumption that the intervention will directly, or indirectly through an intervening construct, influence people's food choice in that setting. Health behavior theories provide an explicit statement of the structural and psychological processes that are hypothesized to regulate behavior (e.g., increasing the cost of high-fat foods will curtail consumption of these foods by making it more aversive or, perhaps, more difficult to purchase them). If theories describe the factors that guide people's behavior and justify how an intervention is designed and implemented, interventionists depend on the quality and predictive value of a theory. What determines a theory's value? From the perspective of a theoretician, a theory's value rests on its ability to provide an accurate account of the factors that regulate people's behavior. [ 17 ] Although investigators may recognize that behavior is affected by factors at different levels of analysis (i.e., biological, psychological, social, environmental), a theory's value is not necessarily predicated on its ability to provide linkages across these levels. Because of this emphasis, theory testing tends to occur in controlled contexts, typically a laboratory setting, that afford the social and behavioral version of a Petrie dish. This approach allows investigators to observe the relation between a given set of constructs with greater precision, but it renders the generalizability and strength of the observed effect difficult to discern. For example, investigators may determine that focusing people's attention on the undesirable aspects of an object increases their interest in avoiding it, but be unable to specify the conditions under which this relation is and is not most likely to obtain. From the perspective of an interventionist, the accuracy of the relations specified in a theory is an important but not sufficient determinant of its value. Interventionists need theories that are accurate and applicable; that specify not only the relation between two constructs, but also whether that relation does or does not change across contexts (e.g., does the impact of risk perceptions on behavior differ whether one is examining decisions to test for radon or to start smoking?). Given a set of a factors hypothesized to regulate people's behavior, interventionists need to be able to discern which of these factors are the most appropriate targets for intervention. In fact, a common complaint regarding theories is that they are not useful (See Jeffery, this issue). A theory may specify a host of factors that regulate a person's behavior, but in the absence of information regarding the relative importance of each factor leave an interventionist unsure as to where to direct her or his resources. For example, the Theory of Planned Behavior [ 18 ] and Theory of Reasoned Action [ 19 ] propose that people's attitudes toward the behavior and their perceived subjective norm regarding the behavior are critical determinants of behavior (albeit mediated by behavioral intention), but the relative contribution of these constructs is allowed to fluctuate from setting to setting. In any given context, it is unclear how to determine a priori which set of constructs should be prioritized as a target for intervention. The interest interventionists have shown in stage-based models of health behavior may reflect the fact that the models attempt to specify the conditions under which specific constructs affect behavioral decisions. [ 8 ] Little guidance is also given as to how or even whether critical constructs can be manipulated. For example, my colleagues and I have proposed that satisfaction with the outcomes afforded by a pattern of behavior is a critical determinant of behavioral maintenance. [ 20 , 21 ] Claims such as this are typically predicated on evidence that measures of a construct, in this case satisfaction, uniquely predict a behavioral outcome. Yet, the observation that someone who is satisfied is more likely to sustain a pattern of behavior does not indicate what causes someone to be satisfied and, thus, little guidance is given as to what can be done to heighten the satisfaction people derive from changes in their behavior. In the absence of this type of information, interventionists may find little difference between developing intervention strategies that are or are not grounded in a health behavior theory. In fact, given these practical needs, it is not be surprising that interventionists are more likely to rely on health behavior theories (e.g., Social Cognitive Theory [ 22 ]) that specify the determinants of its primary constructs and thus provide guidance as to how to construct an intervention protocol. Breakdown in the evolution of health behavior theories If the design and implementation of intervention strategies rely on assumptions regarding the factors that regulate people's behavior, why haven't current theories of health behavior evolved in ways that would enable them to more effectively guide intervention development? I believe the critical problem is that there has been a breakdown in the relation between basic and applied scientists who study health behavior. [ 23 ] As scholars such as Kurt Lewin [ 24 ] have asserted, the development and specification of theories of human behavior depend upon an iterative series of research activities in which theoretical principles initially formulated by basic behavioral scientists are tested and evaluated by applied behavior scientists. These tests provide critical information that enables basic scientists to revise, refine, or reject their initial principles. Moreover, an applied setting can afford investigators the opportunity to assess the relative impact of different processes hypothesized to regulate people's behavior. It is through this on-going cycle of specification, application, and evaluation that accurate and applicable theoretical models arise. To the extent that behavioral theories are not tested in complex social settings such as those afforded by interventions to change health practices, the process by which theories develop is curtailed. Because the manner in which a theory is specified reflects, in part, the contexts in which it has been operationalized and tested, theories that are tested primarily in tightly controlled laboratory settings will likely be characterized by a rich description of the myriad of factors that could affect people's behavioral choices. The laboratory setting allows investigators to minimize noise and potential confounding or moderating factors and thus optimizes their ability to detect processes that can affect people's behavior without determining whether, in a more complex setting, they do affect behavior. [ 17 ] Thus, in the absence of initiatives that empirically test theoretical principles in complex social environments, investigators run the risk of developing a "hot house" theory of health behavior that has limited practical value. Interventions afford an invaluable opportunity to discern the context dependence of causal relations that have been revealed in the laboratory. Some factors may be shown to always be critical, whereas others may be critical only under certain conditions. [ 25 ] For example, self-efficacy may be a critical determinant of the decision to initiate a new pattern of behavior, but have a limited impact on the decision to maintain that behavior over time. [ 21 ] It is critical to understand that restricting the conditions under which a construct affects behavior does not mean that a given factor is not important. Information that would help delimit these conditions would enable theorists to develop more precise models. The case for why interventions should be more receptive to theory There are two sets of reasons why we must take better advantage of the opportunities interventions provide to implement and test theories of health behavior. One set focuses on what theory can do to improve the implementation and evaluation of an intervention, whereas the other set focuses on how interventions can be used to improve the accuracy and quality of prevailing health behavior theories. First, by grounding their work on theoretical principles regarding processes that regulate people's behavior, investigators can readily specify the critical assumptions that underlie their intervention protocol. These formal statements of cause and effect relations not only provide a clear justification for the proposed research activities (i.e., why an investigator believes a given intervention strategy will be effective), but also increase the likelihood that the proposed methodology will allow the investigator to detect whether and why the intervention had its intended effect. [ 10 , 11 ] When faced with unambiguous evidence of a successful intervention effect, investigators might be able to move forward without knowing why the intervention was effective. However, more often than not, investigators are faced with the task of determining why an intervention failed to produced the desired effect or why it worked under a limited set of conditions. An a priori set of theoretical principles can provide an important conceptual and analytic framework for determining why an intervention was ineffective. In particular, it increases the likelihood that investigators have not only identified the constructs that may determine whether an intervention will prove effective, but also assessed them at the appropriate points in the decision process. The second set of reasons why interventions should take advantage of opportunities to test theories of health behavior is that by providing a context in which some or all of the facets of a theory can be tested, interventionists are in a position to generate evidence that will enhance the accuracy and applicability of theory and thus, over time, improve the quality of the theories to which interventionists can turn. By systematically testing principles specified in health behavior theories, investigators are able to not only verify the accuracy of these predictions, but also develop a better understanding of their practical value. Across studies, evidence should accumulate that will allow investigators to differentiate between factors that should and should not be targeted for intervention. Because current theories of health behavior often provide a list of factors that may affect behavior, the set of potential mediating variables suggested by a theory may pose a daunting if not untenable measurement burden. However, the implementation of consistent and methodologically sound assessment of these factors should provide the empirical evidence needed to constrain and prioritize the variables on that list. The characteristics of intervention strategies that prove to be effective should also provide investigators with a better understanding of the determinants of a given construct. As was previously mentioned, theories may propose that a construct (e.g., satisfaction) is a critical determinant of decisions to maintain a new pattern of behavior, but provide limited guidance as to how to alter people's standing on that construct (e.g., how to help feel satisfied with the outcomes afforded by their new behavior). [ 21 ] An intervention protocol that is shown to successfully heighten people's satisfaction with process and outcomes associated with weight loss not only has clear practical value, but also can shed light on the process by which people determine whether they are satisfied with their experiences. If theorists can develop a more detailed account of the processes that shape the primary constructs identified in a health behavior theory, interventionists will find that theories can provide a more useful set of guidelines for how to develop strategies to target these constructs. Testing theoretical principles across a diverse array of settings and populations will also enable investigators to better specify the scope of a theory. Although interventions provide a wonderful opportunity to test theoretical principles in diverse samples and settings, formal and appropriately powered tests of moderators can put a considerable strain on sample size and resources. However, if investigators have appropriately assessed the critical constructs, systematic comparisons can be drawn across studies that taken together have tested a theoretical principle across a range of settings or people. The increase in public access to data sets should facilitate opportunities for this type of comparisons. With the information that is gleaned from these types of activities, it should be easier to determine which moderators are worth testing in a single, appropriately powered study design. The identification of situational or personal factors that moderate the impact of a theoretical principal can be indicative of a number of different scenarios. For example, what might one conclude if an intervention that promoted the health benefits of eating a balanced diet altered the eating habits of college students but not those of high school students? It could indicate that health benefits do not affect what high school students choose to eat. Alternatively, it might be that high school students are responsive to perceptions of the health benefits afforded by a balanced diet, but that other factors (e.g., control over access to food) preclude them from acting on those beliefs. The practical and theoretical conclusions that can be drawn from the identification of moderating factors are dramatically increased if investigators can identify the causal processes that underlie the observed impact of the moderator. In particular, can investigators discern whether the moderated effect was obtained because the moderator altered the ability of the intervention strategy to change the proposed mediating construct (e.g., the intervention raised perceptions of the health benefits held by college but not high school students) or because it altered the effect the mediator has on the primary outcome measure (e.g., perceived health benefits predicted the eating habits of college but not high school students)? Greater attention to the causal processes invoked by a moderator may also help investigators grapple with the daunting number of potential moderators. It is quite possible that moderators that differ at the level of description (e.g., gender, ethnicity) can be accounted for by the same underlying process. Finally, it is important to recognize that progress in theory development can arise from the failure to obtain evidence in support of a specific prediction. Empirical evidence that provides investigators with a better sense of the potential factors that do not affect health practices will allow them to reduce the number of constructs (and, in time, theories) invoked to predict and explain health behavior. What can be done to make interventions more theory-friendly? If one assumes that there is interest in rendering interventions more receptive to theory-testing, what can be done to enhance an intervention's ability to assess principles derived from current health behavior theories? One issue is the appropriate evaluation of the critical manipulation(s) imbedded in the intervention. Any conclusions that can be drawn from the intervention, regardless of whether it reveals the predicted pattern of results, is predicated on the success with which the independent variable was manipulated. To this end, investigators need to at least consider assessing several constructs: the degree to which the intervention was implemented (e.g., did the interventionists consistently provide participants with the intervention exercises?), the degree to which participants correctly identified the emphasis of the intervention (e.g., did participants assigned to the optimistic outcome condition report their was a greater emphasis on favorable outcomes than did those assigned to the control condition?), and finally the degree to which the intervention altered the targeted set of opportunities, thoughts or feelings (e.g., did those assigned to the optimistic outcome condition develop more favorable expectations regarding the benefits afforded by behavior change than did those assigned to the control condition?). Although it is important that interventionists explicitly specify the constructs that determine the influence of the intervention on participant behavior, the quality of the evidence that can be gathered depends on the assessment procedures that are utilized. The persuasiveness of any claims regarding the importance (or lack of importance) of a particular construct is contingent on the use of measures that have been shown to be reliable and valid. Given that many of the constructs specified in theories of health behavior are conceptually similar, it is difficult to draw strong conclusions regarding the specific contributions of different variables in the absence of well-designed measures. [ 26 , 27 ] In addition, the inclusion of a pool of potential mediators enables the investigators to make stronger claims as he or she can demonstrate that not only does the construct specified in the model serve as a mediator but that other factors do not operate as mediators. Adequately testing basic principles also depends on a well-timed assessment schedule. Assessments are often too infrequent to detect meaningful changes on the construct. This is particularly true if the constructs of interest are psychological states that both affect and are affected by behavioral practices. However, specifying the optimal time to assess the primary constructs can be difficult. To the extent that one wants to determine whether an intervention strategy (e.g., a tailored message about dietary changes) alters the predicted mediating variable (e.g., willingness to modify one's diet), one might consider minimizing the length of time between the delivery of the intervention and the assessment of the mediator. However, at the same time, interest in the association between the hypothesized mediator and the outcome variable (e.g., change in diet) would also benefit from a shorter window of time between the two assessments. In many cases, the length of these two time windows are inversely related to each other and thus efforts to improve the chance of detecting one relation may hinder effects to detect the other. Of course, there are practical constraints on an investigator's ability to adequately assess constructs. What is needed is for investigators to take advantage of the measurement and testing opportunities when they do arise. Although what can be concluded from any single assessment effort may be limited, the cumulative impact of well designed tests of a theoretical principle can be substantial. If investigators consistently wait for another time or another investigator to conduct the relevant assessments, innovations in theory and practice will continue to be slow. As interventionists specify the degree to which a given study can test all or a facet of a given theory, they are more likely to articulate the contribution a proposed study could make to the empirical literature. This process not only makes the justification for the intervention clear, but also improves the likelihood that investigators will recognize when their and their colleagues' efforts have focused consistently on a single or limited aspect of a given theory. Research activities motivated by the Transtheoretical Model [ 28 ] provide an excellent example of a domain where researchers have consistently relied on a limited number of methodological strategies and thus, despite an enormous amount of research activity, provided a very narrow test of the theory. [ 8 ] The commitment of time and effort to using interventions to test theoretical principles will in the end be for naught, if there is not an equal commitment to the dissemination of the findings generated by these activities. In particular, investigators who are engaged in the development of health behavior theories must take advantage of the information afforded by intervention activities and demonstrate that they are responsive to this information as they refine and revise their theories. Enhanced communication should also provide an opportunity for basic and applied behavioral scientists to recognize the strengths and weakness of current theories of health behavior and thus help formulate a fuller understanding of what needs to be done to improve the quality of our theories. Summary With an eye toward the future Although Lewin may have been right that there is "nothing more practical than a good theory" (p.169; [ 24 ]), his dictum rests on the assumption that good theories are available to address practical problems. The development of "good" theories – that is, theories that are both accurate and applicable – has been hindered by a breakdown in the on-going collaboration between basic and applied behavioral scientists. Research and professional activities that are able to foster a stronger sense of interdependence between these two groups are likely to provide a base for collaboration and, in turn, a opportunity for innovation. If critical advances in health behavior theory depend on an iterative process by which theoreticians and interventionists cooperate in the testing and evaluation of theoretical principles, individuals in both camps need to not only recognize the goals and values of each group, but also trust each other's ability to advance our understanding of both theory and practice. Competing Interests None declared.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC514573.xml
509278	Characterization of yeast histone H3-specific type B histone acetyltransferases identifies an ADA2-independent Gcn5p activity	Background The acetylation of the core histone NH 2 -terminal tails is catalyzed by histone acetyltransferases. Histone acetyltransferases can be classified into two distinct groups (type A and B) on the basis of cellular localization and substrate specificity. Type B histone acetyltransferases, originally defined as cytoplasmic enzymes that acetylate free histones, have been proposed to play a role in the assembly of chromatin through the acetylation of newly synthesized histones H3 and H4. To date, the only type B histone acetyltransferase activities identified are specific for histone H4. Results To better understand the role of histone acetylation in the assembly of chromatin structure, we have identified additional type B histone acetyltransferase activities specific for histone H3. One such activity, termed HatB3.1, acetylated histone H3 with a strong preference for free histones relative to chromatin substrates. Deletion of the GCN5 and ADA3 genes resulted in the loss of HatB3.1 activity while deletion of ADA2 had no effect. In addition, Gcn5p and Ada3p co-fractionated with partially purified HatB3.1 activity while Ada2p did not. Conclusions Yeast extracts contain several histone acetyltransferase activities that show a strong preference for free histone H3. One such activity, termed HatB3.1, appears to be a novel Gcn5p-containing complex which does not depend on the presence of Ada2p.	Background Histones H3 and H4 are among the most evolutionarily conserved proteins (>90% identity from yeast→humans) [ 1 ]. Octamers composed of one histone H3/H4 tetramer and two histone H2A/H2B dimers package 146 bp of DNA into the basic repeating subunit of chromatin, the nucleosome [ 1 ]. Hence, as fundamental components of chromatin, these proteins are an integral part of all cellular processes involving chromosomal DNA. The physical characteristics of the histones are precisely regulated in the cell by an elaborate network of post-translational modifications that include acetylation, methylation, phosphorylation, ubiquitination and ADP-ribosylation [ 2 - 4 ]. These modifications are found primarily on the NH 2 -terminal tails of the histones. These domains, which protrude from the core of the nucleosome, are free to interact with, and be acted upon by, the nuclear environment. The past several years has seen the identification of numerous enzymes that are capable of modifying the histones. These enzymes are generally found in large, multi-subunit complexes and have activities that are not only specific for a given histone but are specific for particular amino acid residues within the histone [ 5 , 6 ]. The most well characterized histone modifying enzymes are the histone acetyltransferases (HATs). HATs catalyze the transfer of an acetyl moiety from acetyl-coenzyme A to the ε-amino group of lysine residues in the histone NH 2 -terminal tails. Historically, these enzymes have been classified as either type A or type B, based upon substrate specificity and cellular localization [ 7 ]. Found in the nucleus, type A HATs utilize nucleosomal histones as substrates. A number of Type A HATs have been identified in yeast. These include Gcn5p (SAGA, ADA, SLIK, SALSA and HAT-A2 complexes), Sas2p (SAS complex), Sas3p (NuA3 complex), Esa1p (NuA4 and picNuA4 complexes) and Elp3 (Elongator complex) [ 8 - 22 ]. These enzymes have been characterized primarily in the context of transcriptional activation but are likely to be involved in other chromatin mediated events as well [ 23 , 24 ]. Type B HATs were initially described as cytoplasmic enzymes that acetylate free histones in conjunction with chromatin assembly [ 7 ]. The de novo assembly of chromatin is a complex, multi-step process that occurs most prominently during DNA replication (but also accompanies other cellular processes involving DNA synthesis) [ 25 , 26 ]. Following induction of histone mRNA synthesis, histone proteins are translated in the cytoplasm. For histones H3 and H4, synthesis is rapidly followed by the acetylation of specific lysine residues in their NH 2 -terminal tail domains [ 27 ]. For newly synthesized histone H4, this acetylation occurs on lysine residues at positions 5 and 12 in all eukaryotic organisms examined to date [ 28 , 29 ]. For newly synthesized histone H3, acetylation appears to occur in distinct patterns that can differ from organism to organism [ 28 , 30 , 31 ]. The acetylated H3 and H4 form tetramers that are translocated into the nucleus and loaded onto DNA [ 32 ]. Following completion of the histone octamer by histone H2A/H2B addition, mature chromatin is formed following the deacetylation of histones H3 and H4 [ 33 , 34 ]. In contrast to the type A HATs, only one type B HAT has been characterized to date, Hat1p. Hat1p is an evolutionarily conserved enzyme that specifically acetylates free histone H4 [ 35 - 38 ]. Consistent with its identification as a type B HAT, recombinant yeast Hat1p, as well the Xenopus and Human Hat1p homologs, acetylates both lysine 5 and lysine 12 [ 35 - 39 ]. Hat1p was originally purified from yeast cytoplasmic extracts in a complex with Hat2p, a yeast homolog of the mammalian Rbap46/48 proteins [ 36 , 40 , 41 ]. Subsequent studies have shown that yeast Hat1p, as well as its higher eukaryotic counterparts, can also localize to the nucleus [ 37 , 38 , 42 ]. These results suggest that, while specificity for free histones is a bona fide characteristic, cytoplasmic localization may not be a strict criterion for classification as a type B HAT. Evidence has accumulated indicating that the acetylation of newly synthesized histones H3 and H4 play over-lapping roles in chromatin assembly. While yeast strains carrying a deletion of either the H3 or H4 NH 2 -terminal tail are viable, concomitant deletion of both NH 2 -termini (or combining tail deletions with alterations in specific sites of acetylation) results in a defect in nucleosome assembly and cell death [ 43 , 44 ]. In addition, while deletion of the HAT1 gene produces no observable phenotype, combining a deletion of HAT1 with specific lys→arg mutations in the NH 2 -terminus of histone H3 generates defects in both telomeric silencing and DNA damage repair [ 45 , 46 ]. However, despite the importance of the acetylation of newly synthesized histone H3 in chromatin assembly, there have been no type B histone acetyltransferases described that specifically target histone H3. To identify potential histone H3-specific type B HATs, we have systematically surveyed yeast extracts for candidate activities. Here we detail one such activity, termed HatB3.1. We provide evidence that this is a novel complex that utilizes Gcn5p as its catalytic subunit. Intriguingly, unlike previously identified Gcn5p-containing HAT complexes, HatB3.1 contains Ada3p, but not Ada2p. Results Identification of histone H3-specific type B histone acetyltransferase activities in yeast The highly selective activity of the native Hat1p/Hat2p complex for free versus nucleosomal histone H4 is the primary characteristic that distinguishes this enzyme from the type A histone acetyltransferases [ 35 , 36 ]. Therefore, to identify putative histone H3-specific type B HAT complexes, we systematically surveyed yeast extracts for activities that acetylated free histone H3 but not histone H3 packaged into chromatin. Extracts were prepared from cell cultures grown to mid-log phase to enrich for actively dividing cells, as the most robust period of chromatin assembly occurs during DNA replication. Yeast cell walls were digested with zymolyase and cytosolic extracts were produced by the lysis of the cells in low salt buffer followed by centrifugation to remove nuclei and large cell debris. Hence, this extract contained soluble cytoplasmic proteins as well as proteins loosely associated with the nucleus. The nuclear extract was obtained by incubating the nuclear pellet in buffer containing 1.0 M NaCl to extract proteins that are more tightly associated with the nucleus. It is difficult to reliably detect histone acetyltransferase activities in the relatively crude cytosolic and nuclear extracts. Therefore, to evaluate the intrinsic HAT activities present in each of the extracts, they were fractionated by anion and cation exchange chromatography. Fractions were assayed for HAT activity using 3 H-acetyl Coenzyme A and equivalent amounts of either free histones or chromatin as substrate. Histones were then resolved by SDS-PAGE and acetylated species visualized by fluorography. Fractionation of the cytosolic extract on a DEAE column is shown Figure 1A . As expected, the predominant type B activity present in these preparations was attributable to Hat1p, as indicated by robust, free histone H4 acetylation (Fig. 1A , lanes 28–34). The identity of the Hat1p/Hat2p complex was confirmed by western blot analysis using polyclonal antibodies against both Hat1p and Hat2p (data not shown). Figure 1 Identification of putative histone H3-specific type B histone acetyltransferases. Cytosolic and nuclear extracts were prepared and fractionated as outlined in the flow chart. Inherent HAT activities were identified by assaying column fractions with 3 H-Acetyl-Coenzyme A and equivalent amounts of either free histones or chromatin (as indicated). Reaction products were resolved by 18% SDS-PAGE and visualized by fluorography. The relative migrations of the core histones, as determined from coomassie blue staining, are denoted at the side of each fluorogram. The positions of Hat1p and putative H3-specific type B HATs are indicated by brackets. The cytosolic extract also contained at least two additional HAT activities. The first showed a clear peak that was centered on fraction 14 and acetylated free histones H3, H2B and H4. The activity of this HAT on chromatin was more difficult to determine as the H3 and H4 labeling seen in these fractions does not show a marked peak in fraction 14 and may be due to the leading edge of a HAT activity eluting at higher salt. Therefore, this activity may be a candidate type B HAT. There was also a distinct peak of HAT activity at fractions 18–20. With free histone substrates, this activity primarily acetylated histone H3. However, there was also a coincident peak of chromatin H3 and H4 acetylating activity in these fractions suggesting this activity is likely to be a type A HAT. Figure 2 HatB3.1 is a chromatographically distinct activity. DEAE fractions encompassing HatB3.1 activity were pooled and subjected to further fractionation. Fluorograms of liquid HAT assays, using free histones or chromatin as substrates, representative of gradient eluted Mono Q column fractions, showed that HatB3.1 activity can be separated from the overlapping activities present in the initial DEAE fractionation. Migration of the core histones is indicated. Migration of free histone H3 activity attributable to HatB3.1 is marked with an arrow. DEAE fractionation of the nuclear extract also revealed several distinct HAT activities (Figure 1B ). There were two H4-specific type A HAT activities that peaked at fractions 14 and 18, as indicated by activity on both free and nucleosomal histones. There was also a significant peak of activity that acetylated free histone H3 (there was also slight acetylation of histone H2B that is more easily seen in Figure 3 ) that was coincident with a minor nucleosomal H3 HAT activity (Fig. 1B right panel, lanes 16–34). This activity also partly overlapped the nucleosomal H4 activities. This peak of activity was rather broad and most probably results from the partial overlap of at least two distinct activities. In fact, the separation of these activities was readily apparent in Figures 3 and 4 . The strong overall preference of these activities for free histone H3 makes them good candidates for H3-specific type B HATs. As these are chromatographically distinct activities we have termed them HatB3.1 and HatB3.2 as indicated (Figure 1B , right panel). Figure 3 HatB3.1 activity is dependent upon GCN5. Nuclear extracts, generated as depicted in Figure 1 from the indicated isogenic deletion strains, were fractionated via DEAE anion exchange chromatography. Fluorograms of HAT assays resolved by 18% SDS-PAGE are shown with the migration of the core histones as indicated. Fractions of equivalent conductance are aligned for each strain. Regions containing HatB3.1, HatB3.2 and Hat1p are identified by brackets. Figure 4 Highly purified HatB3.1 contains Gcn5p and Ada3p in a high molecular weight complex. A) Flowchart outlining the partial purification of HatB3.1. B) Fluorograms of liquid HAT assays of the Superose 6 column fractionation of HatB3.1 activity (top 2 panels). Assays used either free histones or chromatin as substrate (as indicated). The relative elution of molecular weight standards is shown along the top, while the migration of histones H3 and H4 is indicated at the right. Column fraction aliquots (15 μL) were also resolved by SDS-PAGE and visualized by silver staining (bottom panel, protein ladder mobility is represented at right). Corresponding fraction numbers for both the fluorograms and silver stained gel are indicated along the bottom. C) Peak HatB3.1 containing Superose 6 fractions, as indicated at top of blots, were resolved by three identical 10% SDS gels, transferred to nitrocellulose and probed with the indicated antibodies (left of blots). Presence of Gcn5p, Ada2p and Ada3p in nuclear extract and/or column fractions was visualized via chemifluoresence. Relative migration of protein standards is shown on the right. Unbound material from the initial DEAE fractionation of the cytosolic and nuclear extracts was analyzed by cation exchange chromatography (carboxymethyl sepharose (CM)). While this fraction from the cytosolic extract appeared inactive, there were several additional HAT activities resolved from the nuclear extract (Figure 1C , data not shown). The presence of these activities in the DEAE flowthrough fraction is not simply due to column overloading as recycling the flowthrough fraction over the DEAE column a second time did not result in significant protein retention. Hence, these activities are chromatographically distinct from those that bind the DEAE resin. Two activities, centered on fractions 22 and 30, acetylated primarily histone H4. These appeared to be typical type A HAT's as they were active on both free histones and chromatin. A broad peak of histone H3-specific activity eluted from the CM column from fraction 10 through fraction 24 (with activity trailing through the remainder of the gradient). Comparison of the free histone and chromatin activities in these fractions suggested that this region of the gradient actually contained overlapping type A and type B activities. There was a distinct peak of free histone H3 acetylating activity centered on fractions 12 – 14 while acetylation of chromatin associated H3 peaked in fraction 16. Hence, the activity in fractions 12–14 is another candidate H3-specific type B HAT (labeled HatB3.3). HatB3.1 is specific for free histone H3 The fractions from the DEAE column that contained the activity that we have termed HatB3.1 modified not only free histone H3 but also free H4. In addition, a low level of nucleosomal H3 activity could also be seen in these fractions. To determine whether these activities were the result of a single enzyme complex or were due to multiple, overlapping complexes, these fractions were pooled, dialyzed and fractionated over a Mono-Q column (Figure 2 ). Inspection of the HAT activity profile of the fractions eluting from the Mono-Q column clearly demonstrated that multiple HAT activities overlapped with HatB3.1 during the initial fractionation of the nuclear extract. The HatB3.1 activity eluted from the Mono-Q column very early in the gradient and appeared to be highly specific for free histone H3. The second activity to elute from the Mono-Q column was specific for chromatin-associated histone H4. The third activity acetylated both free and nucleosomal histones H3, H2B and H4. These results indicated that the acetylation of multiple histones in the DEAE elution profile was the result of at least three overlapping activities and confirmed that HatB3.1 is a chromatographically distinct free histone H3-specific activity. Therefore, HatB3.1 was a good candidate for further characterization. HatB3.1 activity is dependent on GCN5 To gain insight into the identity of the catalytic subunit of HatB3.1, we constructed null mutants for each of the yeast HAT's that have demonstrated histone H3 activity as well as the known type B HAT, HAT1 . Isogenic deletion strains (Δ gcn5 , Δ sas2 , Δ sas3 and Δ hat1 ) were grown and protein extracts prepared exactly as for the wild type strain. Nuclear extracts were again fractionated via DEAE column chromatography and fractions of equivalent conductivity assayed for HAT activity as described above. Parallel comparison of the HAT activity profiles from each strain provided biochemical evidence for the dependency of specific histone acetyltransferase activities on the presence of a particular HAT catalytic subunit (compare Figure 3 with Figure 1B ). While subtle variations in observed specificity and intensity of HAT activity were seen throughout the profiles of the Δ sas2 and Δ sas3 strains, the robust H3 acetylation attributed to the HatB3.1 activity appeared unaffected by deletion of these enzymes (Figure 3 , lanes 26–34). Conversely, HatB3.1 activity was abolished in a Δ gcn5 strain (Figure 3 , lanes 26–34). In addition, the HatB3.2 activity also appeared to be absent in extracts from a gcn5 strain indicating that both of these putative type B HAT activities are dependent on Gcn5p. Additionally, the integrity, in a Δ gcn5 strain, of the overlapping free histone and chromatin (data not shown) activities in this region of the gradient confirmed that HatB3.1 was a chromatographically distinct HAT activity exhibiting specificity for free histone H3. Analysis of the activity profile from nuclear extracts derived from a Δ hat1 strain identified a broad peak of Hat1p dependent activity that spanned fractions ~22–34. Western blot analysis using antibodies against Hat1p and Hat2p confirmed the presence of these proteins in fractions from this region of the gradient from the wild type extract (data not shown). As with the Hat1p-dependent activity in cytosolic extracts, this activity also appeared to be specific for free histone H4. This result confirmed previous observations indicating that Hat1p is localized to both the cytoplasm and the nucleus [ 37 , 42 ]. In addition, the presence of an authentic type B HAT activity in our nuclear extracts validated our use of these extracts for the identification of putative histone H3-specific type B HAT activities. HatB3.1 activity is dependent on ADA3 but not ADA2 There are two proteins, Ada2p and Ada3p, that are components of all known Gcn5p-containing HAT complexes and that are required for the activity of these complexes [ 9 - 11 , 13 , 14 ]. To determine whether the HatB3.1 activity was also dependent on these proteins, nuclear extracts were prepared from isogenic Δ ada2 and Δ ada3 strains and the status of the HatB3.1 activity determined by DEAE chromatography. As shown in Figure 4 , the loss of ADA2 did not affect either the HatB3.1 or HatB3.2 activity but did cause a substantial increase in the free histone H4 specific activity that eluted late in the DEAE gradient. However, the HAT activity profile of the Δada3 extracts was strikingly similar to that seen for the gcn5 extracts with both the HatB3.1 and HatB3.2 activities absent. These results indicated that the HatB3.1 activity was dependent on ADA3 and that Ada2p is either not a component of the HatB3.1 activity or is not required for its stability. Partial purification of HatB3.1 To further characterize HatB3.1, this activity was purified through several chromatographic steps. The purification scheme is diagramed in Figure 5A . HatB3.1 containing fractions from the DEAE column were pooled, dialyzed to a conductivity similar to that of the loading buffer (DN(50)) and the dialysate applied to a cation exchange column (CM sepharose). HAT activity assays indicated that the HatB3.1 activity flowed through the CM sepharose column while bound proteins, resolved by a linear salt gradient, contained co-purifying HAT activities that acetylated both free and nucleosomal, H3 and H4 (data not shown). The presence of HatB3.1 in the CM sepharose flow through also confirmed that HatB3.1 and HatB3.3 were distinct activities. Figure 5 ADA3, but not ADA2, is essential for HatB3.1 activity. Nuclear extracts from isogenic Δada2 and Δada3 strains were prepared and fractionated as previously described for wild type and HAT deletion strains (Figures 1 and 3). Fluorograms reflecting HAT activity assays from fractions of equivalent conductivity from each strain, using free histones, are shown (fraction numbers are displayed at bottom [compare lanes to those in figure 3 as well]). Regions of HatB3.1 and HatB3.2 are highlighted by brackets. The proteins that flowed through the CM sepharose column were applied to a Mono-Q column and then eluted with a linear salt gradient. Fractions containing free histone H3 activity were pooled and concentrated by precipitation with 75% ammonium sulfate. The sample was then fractionated by size exclusion chromatography using a Superose 6 column. As seen in Figure 5B , the HatB3.1 activity peaked at fractions 48–50, indicating that a high molecular weight complex of ~500 kDa was responsible for this activity. The size of HatB3.1 remained stable throughout the course of purification as Superose 6 fractionation of the pooled HatB3.1 activity from the initial DEAE column displayed an identical mass (data not shown). The highly purified HatB3.1 retained its high degree of specificity for free histone versus chromatin substrates. There were also two peaks of free histone H4 specific activity seen in the Superose 6 elution profile. Western blot analysis indicated that Hat1p co-eluted with the low molecular weight species. The second peak of H4 activity co-purified with HatB3.1. Whether this acetylation of histone H4 was the result of a weak specificity of HatB3.1 for H4 or due to a second, co-eluting, HAT activity has not been resolved. Gcn5p and Ada3p, but not Ada2p, co-purified with the HatB3.1 activity The absence of HatB3.1 activity in extracts from a Δgcn5 strain indicated that HatB3.1 was dependent on Gcn5p, either indirectly via Gcn5p-mediated transcriptional regulation, or directly, as its catalytic subunit. While the HatB3.1 activity was highly purified relative to the initial nuclear extract, the peak Superose 6 fractions were still too complex to allow the definitive identification of specific bands that co-purified with the activity (Figure 5B ). Extensive efforts to purify HatB3.1 to homogeneity have been unsuccessful. To determine whether Gcn5p was likely to be functioning as the catalytic subunit of HatB3.1, fractions across the peak of HatB3.1 activity from the Superose 6 column were probed with anti-Gcn5p antibodies. As seen in Figure 5C , Gcn5p was present in the fractions containing the peak of HatB3.1 activity from the Superose 6 column. This result is consistent with direct association of Gcn5p with the HatB3.1 complex. Duplicate blots were probed with anti-Ada2p and anti-Ada3p antibodies to determine whether these proteins also co-fractionated with the HatB3.1 complex. As expected, both Ada2p and Ada3p are present in the nuclear extracts (Figure 5C ). However, while Ada3p precisely co-purified with Gcn5p and the peak of HatB3.1 activity, Ada2p did not appear to be associated with this complex. The absence of the Ada2p from the peak of HatB3.1 activity is consistent with the observation that HatB3.1 activity is independent of the ADA2 gene and suggests that Ada2p is not a component of the HatB3.1 complex. The absence of an Ada2p signal on the Western blot was not due to problems with sensitivity as comparison of the relative signals of Gcn5p, Ada2p and Ada3p in the nuclear extracts and Superose 6 fractions demonstrated that the presence of Ada2p in the Superose 6 fractions would have been readily apparent. While the HatB3.1 activity is enriched in the Superose 6 peak fractions relative to the original nuclear extract, the amount of Gcn5p and Ada3p present in these fractions is not enriched relative to the nuclear extract due to the fact that these proteins are components of at least five other histone acetyltransferase complexes. Hence, only a fraction of the Gcn5p and Ada3p present in the cell extracts was associated with HatB3.1. Discussion Considerable genetic and biochemical evidence indicates that, in most organisms, newly synthesized histone H3 is acetylated and that this acetylation plays a role in the de novo assembly of chromatin [ 28 , 30 , 31 , 43 - 48 ]. However, the enzymes responsible for this modification have remained elusive. In the present study we have comprehensively surveyed yeast extracts for putative, histone H3-specific, type B histone acetyltransferase activities. At least three candidate activities were identified, HatB3.1, HatB3.2 and HatB3.3. Further characterization of HatB3.1 indicated that this activity is a novel ~500 kDa HAT complex. In addition, our results suggest that Gcn5p and Ada3p are components of this complex but that, contrary to all previously isolated Gcn5p complexes, HatB3.1 is not associated with Ada2p. It does not appear that the HatB3.1 complex is merely an unstable form of one of the previously characterized Gcn5p-containing complexes as the apparent molecular weight of HatB3.1 did not vary during the course of its purification. There have been at least a dozen distinct HAT complexes identified in yeast [ 8 - 22 , 36 , 42 ]. Conservative analysis of our systematic fractionation of yeast cytosolic and nuclear extracts resolved 12 chromatographically separable activities. However, many of these activities were represented by rather broad peaks, likely to be composed of partially overlapping activities that may differentiate upon further purification (as seen in Figure 2 ). While many of the activities identified here may correspond to previously characterized complexes, it is difficult to determine these relationships, as our initial purification steps differ from those typically used for the isolation of other yeast HAT complexes. In particular, the purification of the SAGA, ADA, SLIK, SALSA, NuA3 and NuA4 complexes start from Ni 2+ -NTA agarose fractionated whole cell extracts, as these enzymes fortuitously bind to this resin [ 9 , 10 , 13 , 14 , 17 , 22 ]. Most histone acetyltransferases have substrate specificities that direct the acetylation of specific residues within one or more of the core histones [ 5 ]. However, these substrate specificities are not fixed and can be altered by the association of the catalytic subunits with different protein complexes [ 19 , 49 ]. The presence of numerous HAT complexes expands the repertoire of modification states that can be generated on the chromatin template. Therefore, as growing evidence indicates that specific cellular processes are associated with precise patterns of histone modification, the presence of multiple HAT complexes in cells is likely to be a reflection of the myriad events that must take place in the context of chromatin [ 50 ]. Despite the importance of histone acetylation in regulating chromatin structure, with the exception of Esa1p, none of the yeast histone acetyltransferases are essential for viability [ 51 , 52 ]. Also, the deletion of most HAT genes results in only relatively mild phenotypes [ 35 , 36 , 53 - 57 ]. One explanation for this observation is that some HATs perform functionally redundant roles in the cell [ 58 , 59 ]. Alternatively, examination of the HAT activity profiles of fractionated extracts derived from HAT deletion strains presented here suggests that there may be mechanisms that can compensate for the lack of one histone acetyltransferase by increasing the activity of other HAT complexes. For example, in a Δ sas2 strain, there is a dramatic increase in an activity present in nuclear extracts that acetylates free histone H4 and which elutes from a DEAE column at a salt concentration similar to that of the nuclear form of Hat1p (Figure 3 ). In addition, deletion of the HAT1 gene causes a large increase in an activity that is coincident with the HatB3.1 activity. These results suggest the possibility that cells may monitor levels of histone modification and adjust specific HAT activities accordingly. HatB3.1 is the third native HAT complex identified from yeast that is only capable of acetylating free histones [ 15 , 36 ]. In addition to the histone H4 specific Hat1p/Hat2p complex, the SAS complex, composed of Sas2p, Sas4p and Sas5p, was recently shown to acetylate free histones H3 and H4. The potential classification of the SAS complex as a type B HAT is supported by the fact that the SAS complex has also been shown to be physically associated with the histone deposition proteins Cac1p and Asf1p [ 16 , 60 , 61 ]. However, the specific target of SAS complex acetylation, histone H4 lysine 16, has not been found to be acetylated in the pool of newly synthesized histones in any organism [ 15 , 30 ]. Therefore, it remains to be determined whether the SAS complex participates in the acetylation of newly synthesized histones H3 and H4 prior to histone deposition or whether it is involved in the post-assembly modification of histones. Gcn5p is the prototypical type A histone acetyltransferase. While rGcn5p is only capable of acetylating free histones under most experimental conditions, it has been identified as the catalytic subunit of five native HAT complexes that acetylate nucleosomal substrates (SAGA, ADA, A2, SLIK and SALSA) [ 9 - 11 , 13 , 14 , 31 , 62 ]. The most straightforward interpretation of the dependence of the HatB3.1 activity on a functional GCN5 gene and the co-elution of Gcn5p with highly purified HatB3.1 is that Gcn5p is also the catalytic subunit of HatB3.1. In the context of the type A HAT complexes, the Ada2p, Ada3p and TAF II 68 proteins have been shown to be important for expanding the substrate specificity of Gcn5p to allow for the acetylation of nucleosomal histones [ 49 , 63 - 68 ]. Hence, the ability of Gcn5p to acetylate histones in chromatin is a property that must be conferred upon it by association with other proteins. The identification of Gcn5p as a component of a type B histone acetyltransferase activity suggests that classification as either type A or type B may not be an inherent property of an enzyme but, rather, may be a function of the association of the enzyme with specific accessory factors. Several properties of HatB3.1 indicate that it is distinct from previously identified Gcn5p-containing complexes. First, HatB3.1 is the only native Gcn5p-containing complex that does not have detectable activity on nucleosomal substrates. Second, the apparent molecular weight of HatB3.1 (~500 kDa), as determined by size exclusion chromatography, is much lower than that of the SAGA, ADA, SALSA and SLIK complexes but is similar to that reported for the A2 complex [ 9 , 13 , 14 , 64 ]. However, unlike HatB3.1, the A2 complex is both dependent upon, and co-purifies with, Ada2p. These results clearly distinguish HatB3.1 as a novel Gcn5p-containing HAT complex [ 64 ]. Ada2p, Ada3p and Gcn5p form a module that provides the catalytic activity to their associated type A HAT complexes [ 5 ]. In these complexes, there does not appear to be any direct physical interaction between Ada3p and Gcn5p but, rather, their association is mediated through Ada2p [ 55 , 67 , 69 , 70 ]. The absence of Ada2p from the HatB3.1 activity suggests that Ada3p and Gcn5p can directly associate under certain circumstances or that another subunit(s) of the HatB3.1 complex can replace the function of Ada2p in bridging the interaction of Ada3p and Gcn5p. The identification of a Gcn5p-containing complex that is independent of Ada2p also suggests that there are cellular processes, such as histone deposition, that are influenced by Gcn5p (and Ada3p) but that do not require Ada2p. However, with the exception of the specific synthetic lethality seen with Δ gcn5 Δ sas3 mutants, deletions of the GCN5 , ADA2 and ADA3 genes have similar in vivo consequences [ 22 , 59 , 71 , 72 ]. The absence of phenotypes unique to Δ gcn5 and Δ ada3 mutants may be the result of the complex functional redundancies observed in the assembly of chromatin. For example, Δ hat1 and Δ hat2 mutants only display phenotypes when combined with mutations in multiple lysine residues in the histone H3 NH 2 -terminal tail [ 45 , 46 ]. Uncovering these redundancies and deciphering the potential role of Gcn5p in the acetylation of newly synthesized histones is likely to require the characterization of the complete set of complexes that display type B histone acetyltransferase activity. Conclusions In conclusion, we have fractionated yeast cytoplasmic and nuclear extracts and resolved several putative histone H3-specific type B histone acetyltransferase activities. One of these activities, HatB3.1, is highly specific for histone H3 that is free in solution. A combination of genetic and biochemical evidence indicates that HatB3.1 is a novel complex that depends on GCN5 and ADA3 but that is independent of ADA2 . Methods Yeast strains UCC1111 was used as the wild type yeast strain that serves as the genetic background for all deletion strains [ 45 ]. Null mutants for GCN5 , SAS3 , SAS2 , ADA2 , ADA3 and HAT1 were constructed using PCR-mediated gene disruption with the HIS3 reporter gene [ 73 ]. Extract preparation Cells were grown to mid-log phase in 1% yeast extract, 2% peptone, 2% glucose and 50 μg/mL ampicillin at 30°C. Cells were harvested at 4000 × g, 10', 4°C and total grams of cells recorded. All buffers contain 1.0 mM PMSF. Spheroplasts were prepared essentially as described previously using 0.25 mg of Zymolyase (U.S. Biologicals) per gram of cells for spheroplasting [ 74 ]. Spheroplasts were burst in 0.5 mL/g cells Lysis Buffer (18% Ficoll 400, 10 mM HEPES [pH 6.0]) followed by dilution in 1.0 mL/g cells Buffer A (50 mM NaCl, 1.0 mM MgCl 2 , 10 mM HEPES [pH 6.0]). Supernatant from a 1500 × g, 15' spin at 4°C was retained as a cytosolic extract. Pelleted material was washed once with Buffer A then resuspended in DN(1000) (DN buffers contain 25 mM Tris [pH 7.5], 10% glycerol, 0.1 mM EDTA and mM [NaCl] listed in parentheses). Supernatant from another 1500 × g spin as above yielded the nuclear extract. This extract was dialyzed O/N at 4°C into DN(0) to a conductivity similar to that of the cytosolic extract. Extracts were cleared by high speed centrifugation (~30,000 × g) prior to their chromatographic fractionation. Extract fractionation All columns were equilibrated with and run using DN Buffers. HPLC (ÄKTA purifier – Pharmacia) was employed for all column runs. Anion and cation exchange chromatography DEAE – Cleared extracts were loaded onto a HiPrep 16/10 DEAE FF column (Pharmacia). Following a 5 C.V. wash with DN(50), proteins were eluted with a linear, 20 C.V., salt gradient from 50 mM to 1.0 M NaCl. A flow rate of 1.0 mL/min. was used and 3.0 mL fractions were collected. CM – Either pooled peak fractions, dialyzed into DN(0) until at similar conductivity as DN(50) start buffer, or Flowthrough from the DEAE were loaded onto a HiPrep 16/10 CM FF column (Pharmacia). The column was washed and proteins eluted as described above. Mono Q – The flowthrough fraction from the CM column was loaded onto a Mono Q HR 5/5 column (Pharmacia). Following a 5 C.V. wash with DN(50), a 20 C.V., linear, salt gradient was employed as above and 0.5 mL fractions were collected. Ammonium sulfate precipitation Peak fractions of HatB3.1 activity from the Mono Q column were pooled and brought to 75% (NH 4 ) 2 SO 4 (0.516 g/mL) over 30' at 4°C. Following an additional 30' equilibration period at 4°C, precipitated protein was pelleted (10,000 × g, 10', 4°C) and resuspended in 300 μL cold, DN(0). Gel filtration chromatography A 250 μL aliquot of resuspended ammonium sulfate precipitate was loaded onto a Superose 6 HR 10/30 column (Pharmacia). The column was equilibrated with and run in DN(350) at a flow rate of 0.3 mL/min. and 0.25 mL fractions were collected. Molecular weight standards (Sigma, MW-GF-1000) were run using the same parameters and 24 μL aliquots of every other fraction run on a 10% SDS-polyacrylamide gel. The elution profile of the MW standards was determined by protein visualization via Coomassie blue staining. Liquid HAT assays Chicken erythrocyte core histones and chromatin were isolated as previously described [ 75 , 76 ]. Typically 10 μL aliquots of column fractions were incubated with 0.1 μM 3 H-Acetyl Coenzyme A (5.50 Ci/mmol, Pharmacia) and ~1.0 mg/mL core histones or chromatin in a final volume of 100 μL at 1X [DN(75)]. 50 μL of each reaction was analyzed for HAT activity via liquid scintillation counting. The remaining assay mixture was brought to 1X [SDS Load Dye] to stop the reaction. In general, aliquots (24 μL) of these remaining assay mixtures were run on 18% SDS-polyacrylamide gels to resolve the histones. Gels were incubated in Autofluor (National Diagnostics), dried down and acetylated histone species visualized via fluorography. Western blot and gel analysis Superose 6 fractions exhibiting HAT B3 activity, as determined above, were run on 10% SDS-polyacrylamide gels and proteins were either visualized by silver staining or transferred to nitrocellulose using a semi-dry transfer apparatus (Biorad). Blots were processed following standard procedures. Goat, polyclonal antibodies against Gcn5p, Ada2p and Ada3p (Santa Cruz Biotechnology, Inc.) were used at 1:100 dilutions in 5% Milk/TBS-T. Donkey, HRP-labeled Anti-Goat IgG secondary antibody (Santa Cruz Biotechnology, Inc.) was used at 1:2500 dilution followed by detection with ECL+Plus (Pharmacia) and visualization via phosphoimager (STORM 860, Pharmacia). Authors' contributions A.R.S. performed all of the experiments presented here and drafted the manuscript. M.R.P. directed the project and edited the manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC509278.xml
546209	Doxycycline-regulated gene expression in the opportunistic fungal pathogen Aspergillus fumigatus	Background Although Aspergillus fumigatus is an important human fungal pathogen there are few expression systems available to study the contribution of specific genes to the growth and virulence of this opportunistic mould. Regulatable promoter systems based upon prokaryotic regulatory elements in the E. coli tetracycline-resistance operon have been successfully used to manipulate gene expression in several organisms, including mice, flies, plants, and yeast. However, the system has not yet been adapted for Aspergillus spp . Results Here we describe the construction of plasmid vectors that can be used to regulate gene expression in A. fumigatus using a simple co-transfection approach. Vectors were generated in which the tetracycline transactivator (tTA) or the reverse tetracycline transactivator (rtTA2 s -M2) are controlled by the A. nidulans gpdA promoter. Dominant selectable cassettes were introduced into each plasmid, allowing for selection following gene transfer into A. fumigatus by incorporating phleomycin or hygromycin into the medium. To model an essential gene under tetracycline regulation, the E. coli hygromycin resistance gene, hph , was placed under the control of seven copies of the TetR binding site ( tetO 7 ) in a plasmid vector and co-transfected into A. fumigatus protoplasts together with one of the two transactivator plasmids. Since the hph gene is essential to A. fumigatus in the presence of hygromycin, resistance to hygromycin was used as a marker of hph reporter gene expression. Transformants were identified in which the expression of tTA conferred hygromycin resistance by activating expression of the tetO 7 - hph reporter gene, and the addition of doxycycline to the medium suppressed hygromycin resistance in a dose-dependent manner. Similarly, transformants were identified in which expression of rtTA2 s -M2 conferred hygromycin resistance only in the presence of doxycycline. The levels of doxycycline required to regulate expression of the tetO 7 - hph reporter gene were within non-toxic ranges for this organism, and low-iron medium was shown to reduce the amount of doxycycline required to accomplish regulation. Conclusions The vectors described in this report provide a new set of options to experimentally manipulate the level of specific gene products in A. fumigatus	Background Aspergillus fumigatus is a saprophytic filamentous fungus that has become the leading mould pathogen in leukemia treatment centers and transplantation units in developed countries, second only to Candida spp . as a cause of systemic mycosis [ 1 ]. Despite some advances in therapy, currently available drugs for the treatment of aspergillosis continue to be hampered by problems with efficacy, toxicity, and the emergence of drug resistance. Moreover, a recent review of the Aspergillus case-fatality rate demonstrated that more than 50% of patients die with, or as a result of, aspergillosis, despite having received the reference standard of therapy [ 2 ]. The continued expansion of the immunosuppressed population emphasizes the need for increased understanding of both the basic biology and virulence of this mould so that more effective antifungal therapies can be developed. The completion of the annotated sequence of the A. fumigatus genome is expected to greatly facilitate efforts to determine the contribution of specific gene products to the virulence of this opportunistic pathogen. Unfortunately, the genetic tractability of A. fumigatus has lagged behind some other fungal systems, particularly in the area of conditional expression systems. Inducible promoter systems have proven to be instrumental for the elucidation of gene function in a number of species, most notably with essential genes. Experimental manipulation of gene expression in A. fumigatus is presently accomplished through the use of DNA cassettes that are introduced into the organism as transgenes [ 3 - 5 ], inserted into specific chromosomal loci [ 3 , 6 ] or expressed from a multi-copy nonintegrating vector [ 7 ]. An inducible expression system based upon the ethanol-inducible alcA promoter from A. nidulans has been successfully used in A. fumigatus [ 8 ]. However, the conditions required to regulate the alcA promoter can have significant effects on the metabolism of the organism and thus remain a concern for many applications, particularly for in vivo studies. The tetracycline operator system has been used to regulate gene expression in a number of species. The system is based upon the E. coli tetracycline-resistance operon, a regulatory unit that detects minute concentrations of tetracycline and mounts an appropriate resistance response. Expression of the operon is controlled by a repressor protein, TetR that binds to operator sequences ( tetO ) in the promoter/enhancer region of the operon and prevents transcription. In the presence of tetracycline TetR is unable to bind tetO , which releases the repression and allows the operon to be expressed. This system has been adapted for experimental gene regulation in eukaryotes by fusing TetR to the VP16 transcriptional activating domain of herpes simplex virus VP16, thereby creating a synthetic tetracycline-regulatable transcriptional activator protein (tTA) that can be used to regulate a gene that is under the control of a tetracycline-responsive promoter (reviewed in [ 9 ] and shown schematically in Fig. 1 ). A tetracycline-regulated promoter is constructed by introducing one or more copies of the tetO sequence upstream of a minimal promoter region and the gene of interest (Fig. 1 ). In the absence of tetracycline, tTA is free to bind to the tetO -promoter and drive the expression of the downstream gene. The addition of tetracycline to the medium prevents tTA from binding the tetO sequences and the promoter is inactive. A variation of this system uses a 'reverse' tetracycline transactivator, rtTA that only binds tetO in the presence of tetracycline. In this case, a gene under tetO control is expressed in the presence of tetracycline, but not in its absence [ 10 ]. The TetR/ tetO system is biologically active in a number of eukaryotes [ 11 - 13 ], including yeasts [ 14 - 17 ], but has not yet been adapted to the filamentous fungi. In this report we demonstrate that the tetracycline-regulated promoter system can be used to manipulate gene expression in Aspergillus fumigatus using a simple co-transfection procedure. Results Vector construction Details on plasmid construction are provided in Methods. The plasmids are shown schematically in Fig. 2 and the individual components are summarized in Table- 1 . Effects of doxycycline on the growth of A. fumigatus Tetracyclines are small lipophilic antibiotics that readily diffuse into eukaryotic cells by passive diffusion. Doxycycline was selected for this study since it has the highest association equilibrium constant to TetR among the common tetracycline derivatives [ 18 ], and has been reported to be most effective in the regulation of tetracycline-regulated promoters in S. cerevisiae [ 19 ]. For the doxycycline system to be effective, the levels of doxycycline required to regulate a tetO promoter must not be within a toxic range for the organism. To determine the range of doxycycline concentrations that are tolerated by A. fumigatus , conidia were spotted onto the center of plates of Aspergillus minimal medium containing 0 – 500 μg/ml of doxycycline and colony diameter was measured with time. Concentrations up to 100 μg/ml had little effect on radial growth rates, all of which were within 5% of each other (Fig. 3 ). However, growth rate was reduced by 16% at 200 μg/ml and by 34% at 500 μg/ml of doxycycline. These results indicate that doxycycline can be used up to 100 μg/ml in minimal medium with no detectable effects on growth rate. Regulated expression of an essential gene by the 'tet-off' system Inducible promoter systems are particularly useful for creating strains that can be inducibly depleted of an essential gene product [ 20 ]. To model an essential gene under tetO control we used heterologous expression of the E. coli hygromycin resistance gene, hph . The hph gene encodes a phosphotransferase that is essential to A. fumigatus in the presence of toxic concentrations of the aminoglycoside antibiotic hygromycin. The hph gene was cloned into a plasmid downstream of a hybrid promoter comprised of seven copies of the TetR binding sequence ( tetO 7 ) linked to a 175 bp minimal gpdA promoter from A. nidulans (p482, Fig. 2 ). The linearized tetO 7 - hph reporter plasmid was co-transfected into A. fumigatus protoplasts together with a linearized plasmid expressing the tetracycline transactivator, tTA (p444, Table- 1 ) and transformants were selected on the basis of their resistance to hygromycin. Although p444 carries the ble gene, phleomycin selection was not included in this experiment. Thirteen hygromycin-resistant colonies were obtained from protoplasts transformed with the tetO 7 - hph reporter construct alone. Since these are integrative plasmids, the observed background colonies are presumed to be a consequence of positional effects at the site of integration, resulting in basal levels of expression of the hph reporter construct. By contrast, 192 colonies were obtained following co-transfection with p444 and the tetO 7 - hph reporter plasmid, suggesting that expression of tTA was driving expression of the tetO 7 - hph transgene and thus conferring hygromycin resistance. Fifty of these hygromycin-resistant colonies were randomly isolated and plated onto secondary hygromycin selection plates in the presence or absence of 100 μg/ml doxycycline. A total of five transformants showed increased hygromycin sensitivity in the presence of doxycycline, two of which were selected for further analysis: one showing marked hygromycin sensitivity in doxycycline (tTA-2) and one showing moderate hygromycin sensitivity (tTA-1). Conidia from each of these transformants were spotted into the center of a plate of minimal medium containing both doxycycline and hygromycin and the radial growth of the colony was monitored with time. The pH of the medium in this experiment was adjusted to 8 in order to maximize the hygromycin toxicity. As shown in Fig. 4A , the tTA-2 transformant showed tight regulation of the phenotype of hygromycin sensitivity. Doxycycline concentrations as low as 30 μg/ml completely arrested growth, indicating that a concentration of doxycycline that is inert to the growth of A. fumigatus (Fig. 3 ) can be used to modulate expression of an essential gene under tetO control in this fungus. Importantly, concentrations of doxycycline below 30 μg/ml could be used to manipulate the degree of hygromycin resistance; at 5 μg/ml and 2 μg/ml of doxycycline, the radial growth rate of the organism was reduced by 68% and 55%, respectively (data not shown). Higher levels of doxycycline were required to suppress the growth of the tTA-1 transformant on hygromycin medium (Fig. 4A ). Northern blot analysis showed that the tTA-1 strain expressed about 5-fold more hph RNA than tTA-2 (Fig. 4B ), which was consistent with the fact that tTA-1 grew faster than tTA-2 in the presence of the same concentration of hygromycin (Fig. 4A , compare tTA-1 and tTA-2, no doxycycline). The doublet shown in Fig. 4B was occasionally seen on Northern blots hybridized to the hph probe and is presumed to represent alternative splicing of the primary hph transcript. The higher levels of hph RNA in the tTA-1 strain could be due to a combination of increased tTA expression (which would be expected to be susceptible to doxycycline regulation) and/or basal expression from one or more integrated copies of the tetO 7 - hph reporter gene (which would not be affected by doxycycline). Since there was a clear dose-response effect of doxycycline on hph expression and hygromycin resistant growth in this strain (Fig. 4A and 4B ), it is likely that the two strains differ primarily in the amount of tTA that they express. Although Northern blot analysis showed barely detectable levels of tTA in either strain (data not shown), undetectable levels of tTA have been reported in other applications of the tetracycline regulatory system and are thought to be due to the toxic effects of overexpression [ 21 ]. Since very low levels of tTA protein are actually required to regulate a tetO promoter [ 21 ], even a small difference in tTA expression level that is beyond the limit of detection of a Northern blot could influence the amount of doxycycline required to suppress tTA activity in this transformant. Regulated expression of an essential gene by the 'tet-on' system A limitation of the tTA-regulated system is that it requires inhibition of transcription rather than activation. To address this, a 'reverse' tTA has been generated (rtTA) that requires interaction of the transactivator with tetracyclines before tetO binding can occur, a system that is referred to as 'tet-on' [ 10 ]. Unfortunately, the mutations that reverse the response to doxycycline also reduce binding affinity for doxycycline ten-fold, thus requiring higher levels of doxycycline for maximal induction. Since there may be adverse effects associated with high doxycycline concentration in A. fumigatus under some conditions [ 22 ], we chose a derivative of rtTA that contains additional mutations that restore binding affinity for doxycycline [ 23 ]. One particular variant, rtTA2 S -M2, also contains a multimerized minimal VP16 activation domain to enhance transcriptional activity, and its sequence has been manipulated to optimize expression in eukaryotic cells [ 23 ]. Using the same co-transfection approach used for the tTA system, the tetO 7 - hph reporter (p500, Fig. 2 ) was co-transfected into A. fumigatus protoplasts together with a linearized plasmid that expresses rtTA2 S -M2 (p474, Fig. 2 ) and the transformants were plated onto medium containing both hygromycin and doxycycline. In this experiment, a modified tetO 7 - hph reporter was used in which a 280 bp terminator sequence from the A. fumigatus cgrA gene [ 29 ] was inserted upstream of the tetO 7 repeats to minimize read-through from flanking sequences (p500). Doxycycline was incorporated into the medium at 100 μg/ml to ensure that the tetO 7 - hph transgene would be expressed at sufficient levels to protect against hygromycin toxicity. Approximately 15% of 27 hygromycin resistant colonies showed reduced growth when shifted to hygromycin medium without doxycycline, one of which was selected for further analysis. As shown in Fig. 5 , the inability of this transformant to grow in the presence of hygromycin was restored by the incorporation of as little as 5 μg/ml of doxycycline into the medium, indicating that low levels of doxycycline are biologically active as regulators of the tetO promoter in A. fumigatus . A further increase in hygromycin resistance was achieved at 15 μg/ml of doxycycline, but concentrations above 15 μg/ml had no additional effect. Northern blots analysis confirmed that the levels of hph RNA in the rtTA transformant were increased by the addition of doxycycline to the medium (Fig. 5 ). When hybridization intensity was normalized to SYBR-green II-stained rRNA bands by phosphorimager analysis, the levels of hph expression in the presence of both concentrations of doxycycline (Fig. 5 ) were thirty-fold greater than in the absence of added doxycycline. Doxycycline-regulation is enhanced by low- iron medium A recent report has shown that iron blocks the accumulation and activity of tetracyclines in bacteria [ 24 ]. Since iron is a standard component of Aspergillus minimal medium, its presence may limit the efficiency of doxycycline-mediated gene regulation, particularly if transcriptional modulators with lower affinity for doxycycline are used. Fig. 6 shows the effects of lowering the iron concentration on doxycycline-mediated suppression of the tetO 7 - hph transgene in the tTA-1 clone showed in Fig. 4 . In comparison to standard minimal medium, where 200 μg/ml of doxycycline was required to reduce expression in this strain (Fig. 4A and 4B ), only 5 μg/ml was required in medium containing one tenth the normal concentration of FePO 4 ·4H 2 0 (Fig. 6 ). This indicates that iron may also impair the accumulation of doxycycline in A. fumigatus and that the choice of medium could have significant effects on doxycycline-mediated gene regulation. Wild type A. fumigatus showed no reduction in radial growth rate on this low-iron minimal medium (data not shown). Discussion The tetracycline-inducible method of gene regulation has become one of the most popular tools to manipulate gene expression in eukaryotes [ 25 ]. The efficacy of the system is attributed to the use of prokaryotic regulatory elements that respond to low concentrations of tetracyclines without affecting eukaryotic physiology, allowing control of gene expression without the concern for pleiotropic effects mediated by the effector. Although widely used in higher eukaryotes, including the model yeast S. cerevisiae [ 19 ], the system has not yet been reported in filamentous fungi. Candida albicans and C. glabrata are the only pathogenic fungi in which the system has been successfully applied thus far, however neither of these studies used the tetR -VP16 fusions upon which the tTA and rtTA systems are based [ 15 - 17 ]. In this study we show that both the tet-off (tTA) and tet-on (rtTA) systems can be used to regulate the expression of a hygromycin resistance reporter gene in A. fumigatus . Since the hph gene is essential in the presence of toxic levels of hygromycin, the ability to control hygromycin resistance by modulating the levels of hph transcription validates the system as a tool for analysis of essential genes in A. fumigatus . In the tTA system we found that individual transformants varied in the amount of doxycycline that was necessary to regulate expression of the tetO 7 - hph reporter gene. Since doxycycline prevents the tTA protein from binding to the tetO sequence, this is most likely due to variability in the amount of tTA protein that is expressed in each transformant. A limitation of the tTA approach described here is that the majority of the hygromycin-resistant transformants from the tTA/ tetO 7 - hph co-transfection were not susceptible to regulation by doxycycline. This may be due in part to leaky expression of the tetO 7 - hph reporter, caused by enhancers in the proximity of the integration site [ 21 , 25 ]. A second possibility is that the levels of tTA coming from the gpdA promoter used in this study were too high to be removed by non toxic concentrations of doxycycline. Since lower levels of tTA expression are more readily suppressed by doxycycline, it is conceivable that a weaker promoter used to drive tTA would increase the frequency with which doxycycline-regulatable transformants can be isolated. Lower levels of tTA expression could also be accomplished by using a shorter segment of the gpdA promoter used in this study. The ability to quantitatively control expression from the tetO 7 - hph reporter gene was also observed in a strain expressing the reverse transactivator, rtTA. Concentrations of doxycycline from 2 μg/ml to 15 μg/ml gave a graded response of hygromycin resistance, indicating that A. fumigatus is responsive to concentrations of doxycycline that are similarly effective in S. cerevisiae [ 19 ] and C. albicans [ 15 ]. Moreover, this level of sensitivity falls within the range of doxycycline concentrations that can be achieved in mouse tissues [ 15 , 16 ], raising the possibility of using this system to modulate the expression of virulence-related genes in pathogenesis studies on A. fumigatus . Only 15% of the hygromycin-resistant colonies from an rtTA/ tetO 7 - hph co-transfection showed doxycycline-dependent hygromycin resistance however, suggesting that some of the hygromycin resistance was due to leaky expression of the tetO 7 - hph gene. Leakage of tetO 7 -regulated genes has been described in other systems, and is attributed to enhancers located in the proximity of the integration site that increase expression of the tetO -linked gene [ 21 , 25 ]. This type of problem will affect tetO 7 -controlled genes regardless of whether they are integrated randomly in the genome or targeted to specific loci. Conclusions This report establishes the utility of the tetracycline-regulated system as an approach to regulate gene expression in A. fumigatus . A limitation of the system was that only 10–15% of the transformants could be regulated by doxycycline, either when tTA or rtTA were used, emphasizing the need to screen for regulatable transformants. A recent approach to limit the problem of leakiness of a tetO -driven gene is the use of trans-silencer proteins comprised of fusions between tetR and a transcriptional silencing domain [ 26 , 27 ]. It is conceivable that the incorporation of a synthetic A. fumigatus -derived trans-silencer protein into the co-transfection approach described in this study would improve the efficiency of the system. Methods Vector construction All vectors are based on the pBluescript plasmid (Stratagene) and were linearized prior to transfection. PCR amplification of components were performed using standard amplification protocols using PfuTurbo DNA polymerase (Stratagene). Hph Reporter Constructs (p482 and p500) A segment containing seven copies of the tet operator sequence ( tetO 7 ) was PCR amplified from pUHD10-3 [ 12 ] with the forward primer 5'- aagctt gcgtatcacgaggccctttc and the reverse primer 5'- aagctt ctcgacccgggtaccgag (added Hin dIII cloning sites are underlined) and cloned into the Hin dIII site of pBluescript. A 1.6 kb fragment containing a minimal gpdA promoter from A. nidulans (-175 relative to the ATG of the hph open reading frame), the hph gene encoding resistance to hygromycin, and the trpC terminator from A. nidulans , was then PCR amplified from pAN7-1 [ 28 ] with forward primer 5'- gagctc cccatcttcagtatattcatc (added Sst I cloning site underlined) and reverse primer 5'- tctaga tcgcgtggagccaagagcgg (added Xba I cloning site underlined) and cloned downstream of tet0 7 into the Sst I and Xba I sites of the plasmid, creating p482. To minimize read-through from flanking sequences into tet0 7 , a 280 bp segment of the terminator region of A. fumigatus cgrA [ 29 ] was inserted upstream of tet0 7 PCR to create p500. The cgrA terminator was PCR amplified from genomic DNA of A. fumigatus isolate H237 using the forward primer 5' aagctt acagcagaagaatctctc (added Hin dIII cloning site underlined) and reverse primer 5' ctcgag atgattcatgacgtatattc (added Xho I cloning site underlined), cloned into pCR2.1-Topo (Invitrogen), excised with Hin dIII, and inserted upstream of tetO 7 in p482 to create p500. tTA expression vectors (p473, p434, and p444) A segment of the A. nidulans gpdA promoter was amplified from pAN7-1 [ 28 ] (position -679 to -1, with +1 being the start of the hph open reading frame) using the forward primer 5'- aagctt cggagaatatggagctt (added Hin dIII cloning site underlined) and the reverse primer 5'- gaattc ggtgatgtctgctcaag (added Eco RI cloning site underlined) and cloned into pBluescript at the same sites. The tTA gene was then PCR amplified from pUHD15-1 [ 12 ] with the forward primer 5'- gaattc tggcaatgtctagattagataaaag (added Eco RI cloning site underlined) and reverse primer 5'-atcatgtct ggatcc tcgcg (internal Bam HI site underlined) and cloned into the Eco RI and Bam HI sites downstream of the gpdA (-679) promoter. A 280 bp segment of the terminator region of A. fumigatus cgrA [ 29 ] was then amplified from H237 genomic DNA using the forward primer 5'- actagt acagcagaagaatctctc (added Spe I site underlined) and reverse primer 5'- gcggccgc atgattcatgacgtatattc (added Not I site underlined) and inserted into the Spe I and Not I sites downstream of tTA. To introduce phleomycin selection into this construct, a phleomycin resistance cassette containing the A. nidulans gpdA promoter, the Streptoalloteichus hindustanus ble gene encoding resistance to phleomycin, and the S. cerevisiae CYC1 terminator was amplified from pBCphleo (Fungal Genetics Stock Center) using the forward primer 5'-cctcaggcggagaatatggagcttcatcg and the reverse primer 5'-cctcaggaattaaagccttcgagcgtccc. The PCR product was cloned into pCR-Blunt II-TOPO (Invitrogen), excised with Kpn I and Xho I and inserted into the P gpdA -tTA construct to create p444. The phleomycin cassette was excised from p444 with Hin dIII and re-ligated to create p473. To introduce hygromycin selection into p444, the phleomycin cassette was excised with Kpn I and Hin dIII and replaced with a hygromycin resistance cassette (containing the A. nidulans gpdA promoter, the hph gene encoding resistance to hygromycin, and the trpC terminator from A. nidulans ) that was amplified from pAN7-1 [ 28 ] with forward primer 5'- ggtacc cggagaatatggagcttc (added Kpn I cloning site underlined) and reverse primer 5'- aagctt gcttgagagttcaaggaag (added Hind III cloning site underlined) to make p434. rtTA expression vectors The tTA gene was excised from p473 with Eco RI and Bam HI and replaced with an Eco RI- Bam HI fragment containing the rtTA2 s -M2 variant of rTA from pUHrT62-1 (generous gift from C. Berens, Erlangen, FRG) to create p474. To introduce phleomycin resistance into p474, the phleomycin resistance cassette was excised from p444 with Kpn I and Hind III and cloned into the same sites in p474 to create p480. To introduce hygromycin resistance into p474, the hygromycin resistance cassette described in p434 was excised from an unrelated plasmid as a Hin dIII fragment and cloned into the Hin dIII site of p474 to make p502. Strains and culture conditions The A. fumigatus strains used in this study are listed in Table- 1 . The wild-type strain, H237, is a clinical isolate. Conidia were harvested from strains grown on Aspergillus minimal medium plates [ 30 ]. This minimal medium contains 4.5 μM FePO 4 ·4H 2 0. For low-iron minimal medium, the FePO 4 ·4H 2 0 concentration was reduced to 0.45 μM. Plasmids were introduced into A. fumigatus protoplasts as previously described [ 3 ]. Following transformation, protoplasts were plated onto 20 ml of osmotically stabilized minimal medium containing 100 μg/ml doxycycline (for transformations involving rtTA-expressing plasmids) or no added doxycycline (for transformations involving tTA-expressing plasmids). After incubating at room temperature overnight, each plate was overlaid with 10 ml of minimal medium top agar containing 0.5% agar, 1M sorbitol, and 8 mg hygromycin B (Invivogen, San Diego, CA). Doxycycline was also incorporated into the top agar overlay (100 μg/ml) for experiments involving rtTA-expressing plasmids. Colonies arising on these primary plates were transferred onto secondary selection plates containing the same selective agents, and conidia from the secondary plates were replated onto selective medium at low density to isolate colonies derived from single conidia. All subsequent experiments were performed on monoconidial isolates. For co-transfection experiments, 5 μg of the linearized tetO 7 - hph reporter construct was co-transfected with 5 μg of the linearized tTA plasmid (p444), or 50 μg of the linearized rtTA plasmid (p474). For experiments addressing the effects of doxycycline on hygromycin sensitivity, ten thousand conidia were spotted onto the surface of Aspergillus minimal medium agar containing hygromycin and doxycycline at the concentrations specified in the Figure legends. The plates were then incubated at 37°C, and colony diameter was measured with time. Radial growth rates were calculated from the exponential part of the resulting growth curves. Northern blot analysis For analysis of hph gene expression, RNA was isolated from overnight cultures in minimal medium supplemented with the indicated concentrations of doxycycline by crushing in liquid nitrogen and extracting RNA from the crushed mycelium with phenol/chloroform. Twenty micrograms of total RNA were fractionated by formaldehyde gel electrophoresis as previously described [ 20 ], transferred to positively charged nylon membranes (MSI, Inc., Westborough, MA, USA) and hybridized to a 32 P-labeled hph DNA probe under stringent conditions in 50% (v/v) formamide/5XSSC (1X SSC is 0.15 M NaCl/0.015 M Na 3 ·citrate, pH 7.6)/2X Denhardt's solution/10% (w/v) dextran sulfate/1% (w/v) sodium dodecyl sulfate (SDS). The hph probe was an 800 bp Eco RI- Bam HI fragment from pAN7-1 [ 28 ] containing a segment of the hph open reading frame. Hybridization intensity was quantified with a Phosphorimager (Molecular Dynamics) and normalized for differences in gel loading by quantitating the relative levels of SYBR-green II-stained rRNA (Molecular Probes, Inc., Eugene, OR, USA). List of abbreviations tTA tetracycline transactivator rtTA reverse tetracycline transactivator TetR tetracycline repressor tetO TetR binding sequence hph hygromycin resistance gene ble phleomycin resistance gene Authors' contributions KV participated in vector construction, gene transfer into A. fumigatus , screening of transformants and drafting the manuscript. RB participated in plasmid construction. JCR contributed to the planning of the study. DSA conceived of the project and directed its design and execution. All authors have read and approved the final manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC546209.xml
544350	Micronutrient malnutrition and wasting in adults with pulmonary tuberculosis with and without HIV co-infection in Malawi	Background Wasting and micronutrient malnutrition have not been well characterized in adults with pulmonary tuberculosis. We hypothesized that micronutrient malnutrition is associated with wasting and higher plasma human immunodeficiency virus (HIV) load in adults with pulmonary tuberculosis. Methods In a cross-sectional study involving 579 HIV-positive and 222 HIV-negative adults with pulmonary tuberculosis in Zomba, Malawi, anthropometry, plasma HIV load and plasma micronutrient concentrations (retinol, α-tocopherol, carotenoids, zinc, and selenium) were measured. The risk of micronutrient deficiencies was examined at different severity levels of wasting. Results Body mass index (BMI), plasma retinol, carotenoid and selenium concentrations significantly decreased by increasing tertile of plasma HIV load. There were no significant differences in plasma micronutrient concentrations between HIV-negative individuals and HIV-positive individuals who were in the lowest tertile of plasma HIV load. Plasma vitamin A concentrations <0.70 μmol/L occurred in 61%, and zinc and selenium deficiency occurred in 85% and 87% respectively. Wasting, defined as BMI<18.5 was present in 59% of study participants and was independently associated with a higher risk of low carotenoids, and vitamin A and selenium deficiency. Severe wasting, defined as BMI<16.0 showed the strongest associations with deficiencies in vitamin A, selenium and plasma carotenoids. Conclusions These data demonstrate that wasting and higher HIV load in pulmonary tuberculosis are associated with micronutrient malnutrition.	Background Approximately one-third of the world's population is infected with Mycobacterium tuberculosis , and the majority live in less developed countries where human immunodeficiency virus (HIV) infection is spreading rapidly. The World Health Organization (WHO) estimates that the number of new cases of tuberculosis and the proportion with coexisting HIV infection will continue to increase [ 1 ]. Immunosuppression increases the risk of developing clinical tuberculosis, which contributes to the increased prevalence of tuberculosis in association with HIV infection. Malnutrition and wasting are associated with tuberculosis, and co-infection with HIV and tuberculosis may potentially exacerbate the wasting that occurs in tuberculosis or HIV infection alone [ 2 - 5 ]. Micronutrient deficiencies have been described in individuals with tuberculosis [ 6 - 17 ] and in those with HIV infection [ 17 - 23 ]. Several cross-sectional studies suggest that patients with tuberculosis are at high risk of deficiencies of vitamin A [ 7 , 10 - 12 ], thiamin [ 13 ], vitamin B 6 [ 14 ], folate [ 6 , 15 ], vitamin E [ 16 ], and zinc [ 10 ]. Poor selenium status has recently been shown to increase the risk of developing mycobacterial disease among HIV-infected injection drug users [ 24 ], but selenium status among HIV-infected adults with pulmonary tuberculosis has not been well characterized. Selenium plays an important role in the selenoenzyme glutathione peroxidase that protects cells against free radical damage and oxidative stress. The relationship between severity of HIV disease and micronutrient malnutrition needs further elucidation. Such information would help identify subgroups that might benefit the most from nutritional interventions. Plasma HIV load was used as an indicator of severity of HIV disease, as HIV load tends to be higher in more active HIV disease. We hypothesized that wasting in pulmonary tuberculosis is associated with micronutrient malnutrition and that HIV-infected adults with pulmonary tuberculosis who have more active HIV disease, as reflected by higher HIV load, also have more severe micronutrient malnutrition. To test these hypotheses, we conducted a cross-sectional study to examine the relationship between wasting and micronutrient malnutrition in HIV-positive and HIV-negative adults with pulmonary tuberculosis in Zomba, Malawi. Methods The study population consisted of adults who presented with new sputum-positive pulmonary tuberculosis in Zomba Central Hospital between July 1999 and April 2003. Subjects were offered HIV testing and were screened for HIV antibodies after signing a written informed consent form. All subjects were given appropriate pre- and post-test HIV counseling. Subjects commenced treatment after enrollment and received standard short course chemotherapy for tuberculosis as per guidelines of the Malawi National Tuberculosis Program [ 25 ]. Adults with a previous history of treated pulmonary tuberculosis were excluded. Three sputum samples from each subject were examined with Auramine-O dark-fluorescent staining method. Sputum positive pulmonary tuberculosis was considered proven when at least one out of three sputum stains showed acid-fast bacilli. HIV infection was diagnosed on the basis of a positive rapid test (Determine 1/2 Rapid test by Abbott, Abbott Laboratories, Johannesburg, SA) and confirmed by a positive enzyme-linked immunosorbent assay for HIV-1 antibodies (Wellcozyme; Wellcome Diagnostics, Dartford, Kent, UK). Plasma HIV load was measured using quantitative HIV-1 RNA PCR (Roche Amplicor Monitor, version 1.5, Branchburg, NJ, USA) with a sensitivity limit of 400 HIV RNA copies mL. CD4 + lymphocyte counts were not conducted due to limited resources. None of the participants were taking antiretroviral treatment. The protocol was approved by the institutional review boards at the Johns Hopkins School of Medicine (Baltimore, Maryland, USA) and the College of Medicine, University of Malawi (Blantyre, Malawi), with final approval by the Office for Protection from Research Risk of the National Institutes of Health. Nutritional assessment Body weight was determined to the nearest 0.1 kg using an adult balance (Seca 700 balance, Seca Corporation, Hanover, MD, USA), and standing height was determined to the nearest cm. Body mass index (BMI) was calculated as body weight/height 2 . Plasma micronutrient concentrations A venous blood sample was collected by venipuncture (Sarstedt Monovette, Newton, NC). Blood samples were shielded from bright light and immediately aliquoted and stored in cryotubes at -70°C. α-carotene, β-carotene, β-cryptoxanthin, lycopene, lutein, zeaxanthin, retinol, and α-tocopherol concentrations were measured in 100 uL of plasma by high performance liquid chromatography using a modified method from the Nutrition Laboratory, Inorganic Toxicology and Nutrition Branch Division of Laboratory Sciences, National Center of Environmental Health, Centers of Disease Control and Prevention (Rosemary Schleicher, personal communication) [ 27 ]. Total plasma carotenoids were defined as the sum of α-carotene, β-carotene, β-cryptoxanthin, lycopene, lutein and zeaxanthin in μmol/L. Plasma trace element concentrations were measured using a Perkin Elmer model AAnalyst 600 atomic absorption spectrometer equipped with Zeeman background correction, a THGA graphite furnace, and an AS800 auto sampler (Perkin Elmer Corp., Norwalk, CT). Quality control was assessed by repeated analysis of pooled human plasma controls run at the beginning and the end of each analysis. Standard curves were run periodically using standard reference material 986C (National Institute of Standards and Technology, Gaithersburg, MD). Throughout all analyses, the plasma samples were run in a masked fashion. Data and statistical analysis Data and statistical analysis were conducted using SAS 8.01 (SAS Institute Cary, NC, USA) and SPSS 9.0 (SPSS, Inc., Chicago, IL, USA). Comparisons between groups were made using t -tests and nonparametric Mann-Whitney U -tests. Univariate analysis of variance was used to test for linear trends across categories of plasma HIV load and BMI. HIV load was categorized into tertiles. HIV negative subjects were assigned a fourth category of HIV load (category 0) when groups were merged for analysis. Nutritional status was assessed in adults with pulmonary tuberculosis with and without HIV co-infection. Subjects were separated into groups according to their extent of wasting. Mild wasting was defined as BMI 17.0–18.49, moderate wasting as BMI 16.0–16.99, and severe wasting as BMI <16.0, conform the WHO strata for BMI grading of severity of malnutrition [ 27 ]. Plasma retinol <0.70 μmol/L was considered consistent with vitamin A deficiency [ 28 ]. Vitamin E deficiency was defined as plasma α-tocopherol <11.6 μmol/L [ 28 ]. Zinc deficiency was defined as plasma zinc <11.5 μmol/L and selenium deficiency as plasma selenium <0.89 μmol/L [ 28 ]. Because there is no standard cut-off for deficiency of carotenoids, we divided total plasma carotenoids into quartiles, with the lowest quartile considered to be the most consistent with deficiency. To examine the risk of micronutrient deficiencies at different severity level of wasting, logistic regression models were fitted with retinol <0.70, α-tocopherol <11.6, zinc <11.5, selenium <0.89, and the lowest quartile of total carotenoids as the outcome variable. Multivariate logistic regression models were conducted to adjust for sex, age and HIV load. A significance level of P < 0.01 was used in this study. Results The study population consisted of 579 HIV-positive and 222 HIV-negative adults with sputum-positive pulmonary tuberculosis. Among the total study population, 66% (232/352) of male and 77% (347/449) of female participants were HIV-positive. The mean age among all subjects was 33 years (range 18–59 years). The majority of subjects were wasted, as 59% of subjects had a BMI <18.5, 32% of subjects had a BMI <17.0, and 17% of all subjects were severely wasted as defined by BMI<16.0. Plasma retinol concentrations <0.70 μmol/L occurred in 61% of all subjects. Vitamin E, zinc, and selenium deficiency occurred in 13%, 85% and 87% respectively. Table 1 shows characteristics of study participants, such as sex, age, BMI, and plasma carotenoids, retinol, α-tocopherol, zinc and selenium by categories of plasma HIV load. BMI, plasma retinol, total carotenoids and selenium concentrations decreased by increasing plasma HIV load. Age, the proportion of individuals with BMI <18.5, BMI <16.0 and selenium deficiency were increased with increasing plasma HIV load. Plasma α-tocopherol, zinc or the proportion of individuals with vitamin A, vitamin E, or zinc deficiencies were not significantly different across the categories of plasma HIV load. When exploring across the categories separately, there were no significant differences between HIV-negative individuals compared with HIV-positive individuals in the lowest tertile of viral load. Table 2 shows adjusted odds ratios (O.R.) and 95% confidence intervals (C.I.) for independent associations between wasting and micronutrient deficiencies. Wasting defined as BMI<18.5 was associated with vitamin A deficiency, low plasma carotenoids and selenium deficiency. The odds ratio for an independent association with vitamin A deficiency was 2.86 (95% C.I. 2.11–3.89) when adjusted for sex, age, and plasma HIV load. The adjusted odds ratio for an independent association with the lowest quartile of total carotenoids was 2.96 (95% C.I. 1.99–4.44). The adjusted odds ratio for an independent association with selenium deficiency was 1.59 (95% C.I. 1.04–2.43). When separating severity levels of wasting; mild wasting did not show association with deficiencies, moderate wasting was associated with vitamin A deficiency and severe wasting was significantly associated with vitamin A deficiency, low plasma carotenoids and selenium deficiency. (Table 2 ) Wasting was not associated with vitamin E or zinc deficiency. Figures 1 , 2 and 3 show plasma retinol, total plasma carotenoids, and plasma selenium concentrations with 95% C.I by severity of wasting and categories of plasma HIV load. Plasma retinol concentrations significantly decreased with the increase of plasma HIV load among non-wasted adults with pulmonary tuberculosis ( P = 0.004). Total carotenoid concentrations significantly decreased with the increase of plasma HIV load among non-wasted, mildly wasted, moderately wasted and severely wasted adults ( P = 0.0001, P = 0.002, P = 0.001 and P = 0.001, respectively). Selenium concentrations decreased significantly with the increase of plasma HIV load among non-wasted and severely wasted adults with pulmonary tuberculosis ( P = 0.0001 and P = 0.03, respectively). Among the HIV negative adults and those in the 1 st and 2 nd tertile of HIV load, plasma retinol, total carotenoids and selenium concentrations significantly decrease with the increasing severity of wasting. Among those in the 3 rd tertile of HIV load, only plasma retinol concentrations significantly decreased with the increasing severity of wasting. This trend did not reach significance for plasma carotenoid and selenium concentrations. Discussion The present study shows that micronutrient malnutrition and wasting are more severe in adults with pulmonary tuberculosis who have higher plasma HIV load. The association between high plasma HIV load and nutrient deficiencies was strongest for the major plasma carotenoids and selenium. Overall in this study population, both HIV-positive and HIV-negative adults with pulmonary tuberculosis were extremely malnourished as indicated by BMI and plasma micronutrient concentrations. About one-third of the adults in this study had a BMI <17.0, a cut-off that is predictive of mortality in adults co-infected with tuberculosis and HIV [ 29 ]. To our knowledge, this is the first study to demonstrate that selenium status is extremely poor among HIV-infected adults with pulmonary tuberculosis, and that the extent of selenium deficiency is associated with higher plasma HIV load. This observation may be of potential importance because selenium deficiency has been associated with increased mortality during HIV infection [ 30 ], and selenium supplementation for HIV-infected adults has been shown to reduce morbidity [ 31 ]. In the present study, selenium deficiency occurred in 87% of the participants, which, to our knowledge, may be the highest prevalence of selenium deficiency reported in an HIV-infected group of patients. It is unknown whether selenium supplementation will reduce morbidity and mortality among HIV-infected adults with pulmonary tuberculosis. Carotenoids are among the most important dietary antioxidants found in human plasma, and this study shows that poor carotenoid status was associated with higher HIV load and with wasting. Plasma carotenoid concentrations are widely considered to be the most accurate indicator of dietary carotenoid intake [ 32 ]. It is not known whether adults with pulmonary tuberculosis and higher HIV load have lower plasma carotenoid concentrations because of increased oxidative stress, or whether these individuals are sicker and unable to consume enough carotenoid-rich foods. Further studies are needed in the future to address dietary intake of carotenoids in HIV-infected adults with pulmonary tuberculosis. Low BMI is a known risk factor for mortality [ 5 , 29 ], and the present study showed that the risk of micronutrient deficiencies is highest in those with low BMI. HIV-infected adults with wasting and high viral load were at the highest risk of more severe micronutrient malnutrition, suggesting that this subgroup might potentially benefit the greatest from nutritional interventions. The cross sectional design of this study restricts our conclusions and does not provide information on whether poor nutritional status is a predictor of more severe pulmonary tuberculosis. It is unknown whether nutritional interventions will slow progression of disease or reduce wasting associated with morbidity and mortality if added to tuberculosis chemotherapy. Controlled clinical trials currently in progress in developing countries should help provide insight into the role of micronutrient supplementation for HIV-positive and HIV-negative adults with pulmonary tuberculosis. Conclusions The present study shows that micronutrient malnutrition and wasting are more severe in adults with pulmonary tuberculosis who have higher HIV load. The association between high plasma HIV load and nutrient deficiencies was strongest for the major plasma carotenoids and selenium. Further longitudinal investigations are needed to determine whether deficiencies in micronutrients are independent risk factors for increased morbidity and mortality. Competing interests The author(s) declare that they have no competing interests. Authors' contributions Overall guidance and initial study design was provided by RS. MvL has been in charge of the collection and analysis of data and writing of the manuscript. Provision of advice was given by AH, JK, EZ, TC and TT. All authors read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC544350.xml
423136	Virtual Labs: E-Learning for Tomorrow	At Stanford University, the Virtual Labs project takes full advantage of information technology to provide innovative resources for learning and teaching	Because of the explosive growth in our scientific understanding, today's students are required to learn and maintain a rapidly expanding knowledge base. Students are also expected to understand and follow the crossover of information between different disciplines. As a result, they often have to understand the fundamentals of several disciplines, and be able to integrate that knowledge. Students of every discipline are facing these new challenges, and it is clear that today's students are markedly different from those of the past. Influenced by a lifetime surrounded by media, computers, and the Internet, they bring with them different expectations. As educators, we need to meet these expectations in order to motivate students to move forward. And it's not just the student population that is driving change. The National Institutes of Health, which sponsors many biological and medical advances in the United States, has a new initiative called “Digital Biology: The Emerging Paradigm,” whose goal is to merge biomedical computation with biology and medicine over the next ten years. One way to facilitate this movement is to use information technology (IT) as a teaching tool, so that students, in turn, learn how to use IT most effectively. Using IT to Teach IT presents educators and teachers with a unique opportunity to devise innovative methods of teaching. Students today are more likely than ever to use new tools and technologies to advance their understanding of the sciences. Currently, this usage is mainly limited to searching the Web for information. However, computers and the Web can be used for much more—with computers, you can create learning scenarios like virtual patient simulations, and with the Web, these learning resources can be disseminated to the global community. Educators must harness the power of these enabling technologies, which students have already adopted, to create new and more powerful methods of teaching that will better prepare the students for the next phase of their lives. The Virtual Labs Project at SUMMIT (Stanford University Medical Media and Information Technologies) in the Stanford University School of Medicine has been funded by the Howard Hughes Medical Institute (Chevy Chase, Maryland, United States) since 1998. It is an initiative to augment the Department of Biological Sciences and the Program in Human Biology at Stanford University by developing technology-enhanced materials for these curricula. A major goal of the Virtual Labs Project is to increase scientific literacy by using interactive multimedia to teach the fundamental concepts of biology, and to share those resources via the Internet. The Virtual Labs material is currently hosted on a password-protected site and is freely available to interested parties for educational use. A wise individual once said that a picture is worth a thousand words; with Virtual Labs we use not just pictures, but also animations and interactive simulations. Students are able to visualize and interact with dynamic processes in the body. We have developed learning modules in cardiovascular, gastrointestinal, respiratory, renal ( Video 1 ), visual ( Video 2 ), and neurophysiological systems. The concepts in these modules lay the foundation for medicine and for an increasing number of interdisciplinary programs, such as biomedicine, medical informatics, and bioengineering. For example, a medical student learns how the kidney filters blood in order to understand kidney failure in diabetic patients. A bioengineer could apply the same knowledge to build an artificial kidney. The modules are flexible, and the content can be woven together to highlight the intersections of different disciplines. Virtual Labs also strives to make learning science fun. The more engaged the user is, the more likely the learning experience is to be positive. For example, after learning about how the kidney filters blood (see the online link for Figure 1 ) and controls water levels, students apply their new knowledge by playing a simulation game. The goal of the game is to maintain water balance in order to survive on a deserted island, which helps to reinforce conceptual understanding and to ensure that students understand how those concepts fit together. Responses from students have shown that these goals are being met. Over the past four years, our undergraduate and medical students have reported that Virtual Labs was fun, engaging, and that it helped them learn: “Virtual Labs was an excellent resource for the class! [It was] a lot of fun to use and the graphics are awesome.” “It was a great interactive way to reinforce what I already had learned from the book and lectures and I think it really helped me better understand.” Similarly, faculty who have used our material during lectures have found it useful to illustrate concepts with animations. Reaching Beyond the Local Community Local schools and other universities are looking for opportunities to bring IT into their classrooms. Access to resources like Virtual Labs, and expertise on how to develop and integrate multimedia content into curricula, are on the rise. In 2003, the Virtual Labs Project began building a network of collaborators in the community and abroad. Together with H.E.L.P for Kids ( http://www.stanford.edu/group/help ,) we are designing content for the education of local schoolchildren. Abroad, we are working with global partners in Sweden via the Wallenberg Global Learning Network ( http://www.wgln.org ). We have also partnered with the MedFarmDoIT group at Uppsala University in Uppsala, Sweden ( http://doit.medfarm.uu.se/multimedia.html ) to help them with multimedia development and IT integration in the classroom. The MedFarmDoIT group shares our vision of bringing IT into the classroom, and together, we are designing content for their medicine/pharmacy program. The Virtual Labs Project is dedicated to supporting its partners by distributing customized Virtual Labs content and offering consultation or workshops to train teachers and developers. As integrated partners, we can bridge the gap between the physical and information sciences, and in doing so, can improve the learning process of students for years to come. Video 1 The “Big Picture” of the Blood Flow through the Vasculature in the Kidney The rich visuals and moving media of this Virtual Labs animation capture the attention of the students. (The animation can be accessed online on computers with Shockwave by clicking and dragging the file into the browser window. A free version of Shockwave can be downloaded from http://sdc.shockwave.com/shockwave/download .) Video 2 Understanding Center–Surround Receptive Fields in Retinal Neurons The virtual experiment in this Virtual Labs interactive program is similar in design to the receptive field experiments from Hubel and Wiesel in the 1960s. The user places an electrode in the retina to take a recording from a neuron. The user moves a spot of light on the screen and then maps correlating changes in the activity level (using the symbols: + − 0 to indicate the strength of the response). The map reveals a center–surround organization. Each time the user moves the electrode, the size, shape, location, and type of receptive field changes (on-center or off-center), as they would during a real experiment. Supporting questions adapt dynamically to each experimental condition and further encourage the student to answer more conceptual questions. (The interactive program can be accessed online on computers with Shockwave by clicking and dragging the file into the browser window. A free version of Shockwave can be downloaded from http://sdc.shockwave.com/shockwave/download .)	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC423136.xml
546221	Stable expression of constitutively-activated STAT3 in benign prostatic epithelial cells changes their phenotype to that resembling malignant cells	Background Signal transducers and activators of transcription (STATs) are involved in growth regulation of cells. They are usually activated by phosphorylation at specific tyrosine residues. In neoplastic cells, constitutive activation of STATs accompanies growth dysregulation and resistance to apoptosis through changes in gene expression, such as enhanced anti-apoptotic gene expression or reduced pro-apoptotic gene expression. Activated STAT3 is thought to play an important role in prostate cancer (PCA) progression. Because we are interested in how persistently-activated STAT3 changes the cellular phenotype to a malignant one in prostate cancer, we used expression vectors containing a gene for constitutively-activated STAT3, called S3c, into NRP-152 rat and BPH-1 human benign prostatic epithelial cells. Results We observed that prostatic cell lines stably expressing S3c required STAT3 expression for survival, because they became sensitive to antisense oligonucleotide for STAT3. However, S3c-transfected cells were not sensitive to the effects of JAK inhibitors, meaning that STAT3 was constitutively-activated in these transfected cell lines. NRP-152 prostatic epithelial cells lost the requirement for exogenous growth factors. Furthermore, we observed that NRP-152 expressing S3c had enhanced mRNA levels of retinoic acid receptor (RAR)-α, reduced mRNA levels of RAR-β and -γ, while BPH-1 cells transfected with S3c became insensitive to the effects of androgen, and also to the effects of a testosterone antagonist. Both S3c-transfected cell lines grew in soft agar after stable transfection with S3c, however neither S3c-transfected cell line was tumorigenic in severe-combined immunodeficient mice. Conclusions We conclude, based on our findings, that persistently-activated STAT3 is an important molecular marker of prostate cancer, which develops in formerly benign prostate cells and changes their phenotype to one more closely resembling transformed prostate cells. That the S3c-transfected cell lines require the continued expression of S3c demonstrates that a significant phenotypic change occurred in the cells. These conclusions are based on our data with respect to loss of growth factor requirement, loss of androgen response, gain of growth in soft agar, and changes in RAR subunit expression, all of which are consistent with a malignant phenotype in prostate cancer. However, an additional genetic change may be required for S3c-transfected prostate cells to become tumorigenic.	Introduction Signal transducers and activators of gene transcription (STATs) are, as their name suggests, proteins that regulate gene expression by affecting transcription. They are part of the signal transduction pathway used by many growth factors and cytokines, and are activated by phosphorylation of tyrosine and serine residues by up-stream kinases [ 1 ]. For example, signaling by IL-6 and other members of this cytokine family generally induces phosphorylation of STAT3 [ 1 , 2 ]. In the example given in Figure 1 , IL-6-induced binding to its receptor leads to homodimerization of the receptor, which in turn leads to autophosphorylation of the cytosolic domain of gp130; this in turn causes the phosphorylation of one of 3 kinases, JAK1, JAK2, or Tyk 2. The activated up-stream kinase phosphorylates STAT3, which allows for dimerization of STAT3 although this concept is currently being revisited, since it has been shown in hepatic cells under inflammatory stress, there is evidence for STAT3 association on lipid rafts prior to phosphorylation [ 3 , 4 ] in association with chaperone proteins such as Hsp90 (reviewed in [ 5 ]); however only the dimer form of STAT3 can translocate and bind to DNA at specific binding sites, thereby directing transcription of target genes. In benign cells, the signaling by STAT3 is under tight regulation, so that the signal delivered to the cell is transient. However aberrant signaling by STAT3 has been noted in many types of malignancies, such as myeloma, head and neck cancer, breast cancer, and prostate cancer [ 6 - 9 ]. Such persistent signaling by IL-6 leading to aberrant activation of STAT3 is thought to play a role in neoplastic progression of prostate cells [ 10 ]. Importantly, we and others have shown that malignant prostate cells expressing persistently-activated STAT3 become dependent upon this transcription factor for survival, resulting in apoptosis [ 11 - 13 ]. Thus, persistently-activated STAT3 fulfills the criteria of a proto-oncogene [ 14 , 15 ]. Figure 1 An example of cytokine-mediated activation of STAT3. In this example, IL-6-induced binding to its receptor leads to homodimerization of the receptor, which in turn leads to autophosphorylation of the cytosolic domain of gp130; this in turn causes the phosphorylation of one of 3 kinases, JAK1, JAK2, or Tyk 2. The activated up-stream kinase phosphorylates STAT3, which allows for dimerization of STAT3; only the dimer can translocate and dock to DNA at target genes, thereby directing transcription. Prostate cancer (PCA) is the second most frequently diagnosed non-cutaneous malignancy in American males, affecting approximately 35% of them according to recent data [ 16 , 17 ]. This translates into approximately 35,000 deaths last year in the United States alone; 189,000 new cases were diagnosed in 2002 and over 220,000 cases were projected for 2003 [ 18 , 19 ]. Moreover, in a recent report the authors claimed that 30% of male mortality overall may be due to prostate cancer [ 20 ]. For the most effective therapy with the fewest side-effects, a thorough understanding of the genes involved in the neoplastic process is essential. Androgens are known to play a critical role in the tumorigenic process, with activity mediated by the androgen receptor. Initially, prostate cancers are androgen-sensitive (that is, they cease growing when deprived of androgens or when treated with androgen receptor antagonists, such as flutamide or bicalutamide), and therefore most patients respond to androgen ablation therapy. However, there are side-effects to this therapy that make it unpleasant for the patient [ 21 ]. Even with androgen ablation therapy, the disease often recurs and when it does, it usually becomes androgen-insensitive or hormone-refractory [ 22 ]. There is evidence that STAT3 activation via IL-6 plays a role in the conversion of normal prostate cells to prostate cancer cells, and from androgen-responsive to the androgen insensitive phenotype [ 10 , 23 , 24 ]. The progression to androgen-independence has been found to be associated with IL-6, with c-myc expression, and with insulin-like growth factors, all of which can signal through the activation of STAT3 [ 25 - 28 ]. STAT3 is negatively regulated by a retinoid-sensitive protein, GRIM-19, which may explain the positive effects retinoids show against prostate cancer cells in vitro [ 29 - 31 ]. Retinoid therapy for the treatment of prostate cancer is currently being tested, due to the ability of these compounds to rapidly induce apoptosis [ 32 ]. Indeed, the recent addition of Taxotere to the pharmacopeia for prostate cancer may well be due to its demonstrated effect on retinoid receptors [ 33 ]. The regulation of the expression of the 3 retinoid receptors type A (RAR-α, -β, and -γ) in the progession to prostate cancer has been partially addressed by Richter, et al., who showed the differential effects of all- trans retinoic acid in human prostate cancer lines [ 34 , 35 ] To this end we are studying the oncogenic role of STAT3 activation in rat prostate epithelial cell lines NRP-152 [ 36 ] and human benign prostatic hyperplasia line BPH-1 [ 37 , 38 ]. Our main hypothesis is that constitutively-activated STAT3 (cSTAT3) plays an essential role in the development of PCA and the maintenance of the malignant phenotype. Because prostate epithelial cells become hypertrophic, but rarely malignant, they are useful for studying the progression to neoplasia to see how a relatively transformation-resistant cell type becomes neoplastic through cSTAT3. We previously determined that STAT3 was constitutively phosphorylated (hence activated) in malignant NRP-154 but not in NRP-152 cells, even when the NRP-152 cells were treated with testosterone [ 10 ]. We hypothesized that cSTAT3 may account for the tumorigenicity of NRP-154 cells, and therefore may play a determining role in the progression from hyperplasia to neoplasia. To test our hypothesis, we transfected a plasmid containing a mutated gene for STAT3 known as S3c, in which a Cys residue was substituted for an Ala residue, thereby allowing the dimerization of the mutated STAT3, which can then translocate across the nuclear membrane and effect gene transcription in much the same way as the phosphorphylated wild-type STAT3 gene product [ 14 , 39 ] into NRP-152 and BPH-1 cells. We then examined the phenotype of the selected transfected cells after cloning by limit dilution. Our results, indicating that NRP-152 and BPH-1 cells underwent changes in phenotype consistent with that of malignant cells, are presented here. Results Selection of Transfected NRP-152 and BPH-1 Cells Two weeks after transfection with either pIRES or pIRES-S3c and selection with G418, no surviving cells were observed in the wells that received Clonfectin only. Growth of cells was observed in all wells that received either of the plasmids plus Clonfectin. Transfected cells were expanded for further analysis in complete medium. A summary of cells and clones and what their phenotypes were is given in Table 1 . To summarize briefly, since the full results will be discussed in this section, we observed the following changes: Table 1 Summary of transfected cells Growth Factor G418 FLAG EGFP Growth In Cell Plasmid Dependence Sensitivity Epitope Expression Simple Medium NRP-152 none x x - - - 152-pIRES pIRES-EGFP x - - x - 152-S3c pIRES-S3c - - x x x BPH-1 none n/a x - - n/a BPH-pIRES pIRES-EGFP n/a - - x n/a BPH-S3c pIRES-S3c n/a - x x n/a The table summarizes the cells used or made in the experiments described. Cells were transfected with the indicated plasmid, as described in Materials & Methods. G418 was added at 400 μg/ml. Growth factors were those added to 152 medium, but not to 154 medium, as described in Materials and Methods, for NRP-152 cells and transfectants. X = presence of characteristic; --- = absence of characteristic. NRP-152 cells require a variety of growth factors and additives in their medium (see Materials and Methods Section; [ 36 ]); 152-pIRES cells (NRP-152 cells transfected with pIRES-EGFP) required the same medium as NRP-152 cells. But 152-S3c cells grew in DMEM/Ham's F12 supplemented only with 10% newborn calf serum. Moreover, 152-S3c cells expressed EGFP (as did 152-pIRES, which was expected since they were transfected with pIRES-EGFP) and the FLAG epitope, which is part of the S3c gene [ 40 ]. Both 152-pIRES and 152-S3c cells grew in the presence of G418. BPH-1 cells grow in RPMI-1640 supplemented with bovine serum; therefore this line does not have growth factor dependence to begin with. BPH-pIRES and BPH-S3c cells, aside from exhibiting G418 resistance, expressed EGFP, but only BPH-S3c expressed the FLAG epitope of the S3c gene. The evidence for these observations given in Table 1 is presented in the rest of this section. Expression of FLAG and EGFP in 152-S3c and BPH-S3c Cells Was Observed After Transfection and Selection with Antibiotics After no viable cells were observed following antibiotic treatment, we analyzed transfected cells for the presence of the markers flanking the S3c gene on the plasmids used, FLAG and EGFP. The analyses were done by flow cytometry on a FACScan; also by Western blot using specific Abs, and the results are presented in Figure 2 . In Panels A through D, the mean fluorescence intensities of representative clones of 152-S3c and BPH-S3c cells stained with monoclonal Ab to FLAG plus fluorescent goat anti-mouse F(ab 2 )', as well as the enhanced green fluorescent protein fluorescence intensities of transfected cells, are shown. Panel A displays the anti-FLAG fluorescence intensity of 1 representative clone of 152-S3c (thin line) compared to untransfected NRP-152 cells (thick line); approximately 95% of the 152-S3c cells stained with the anti-FLAG antibody. Similary, Panel B shows the fluorescence intensity of anti-FLAG-stained BPH-1 cells (thin line) compared to anti-FLAG-stained BPH-S3c clone (thick line), where approximately 76% of the BPH-S3c cells stained with the anti-FLAG antibody. Panels C and D display the EGFP fluorescence for clones of 152-S3c and BPH-S3c cells, compared with untransfected cells, respectively. In Panel C , the thick line shows the fluorescence intensity of EGFP in 152-S3c and the thin line shows the lack of EGFP fluorescence in the untransfected NRP-152 cells. Approximately 67% of the 152-S3c cells showed EGFP fluorescence. In Panel D , the thin line shows the EGFP fluorescence intensity of BPH-S3c cells, while the thick line shows it for untransfected BPH-1 cells. Approximately 45% of the BPH-S3c cells showed fluorescence due to EGFP. We concluded that in addition to antibiotic resistance, the transfected cells expressed markers flanking the S3c gene, and therefore we could attribute any change in phenotype of the cells to the expression of the S3c, in comparison to the vector-transfected cells. Panel E shows the results of immunoprecipitation with anti-FLAG Ab, followed by Western blot to detect EGFP. We used anti-FLAG Ab for the immunoprecipitation because (1) a S3c-specific Ab is not available, and (2) because all cells express STAT3. Thus, because expression of FLAG equates with expression of S3c specifically, immunoprecipitating with anti-FLAG would reveal the S3c-expressing cells. As seen in Figure 2E , the bands corresponding to 27 kD EGFP are visible only in the lanes from 152-S3c and BPH-S3c cells, while no EGFP bands are visible in the bands from the parental lines NRP-152 and BPH-1 cells. Since the EGFP gene is 3' to the S3c gene in the pIRES-S3c plasmid we constructed (the plasmid codes for a bicistronic message with 1 promoter for EGFP and S3c), these results confirm the flow cytometry data shown in Panels A through D. Figure 2 FLAG and EGFP expression in representative NRP-152 and BPH-1 clones transfected with either pBABE-S3c or pIRES-S3c. NRP-152 and BPH-1 cells were transfected with pBABE-S3c or pIRES-S3c, which bear the FLAG epitope on the S3c gene. Clones were derived by limit dilution, as described in Materials & Methods. Panels A–D: In all histograms, the marker M1 sets the region of positively fluorescent cells for determining the percent positive cells. Panels A & B: Fixed cells were permeabilized and stained with anti-FLAG M1 Ab (Sigma), as described in Materials & Methods. Controls for staining were included, as described. Panel A: Transfected NRP-152 cells. Thin line = 152-S3c; thick line = NRP-152. Panel B: Transfected BPH-1 cells. Thin line = BPH-1; thick line = BPH-S3c. Panels C & D : NRP-152 and BPH-1 cells transfected with pIRES-S3c were analyzed for EGFP fluorescence, following selection. Panel C: Transfected NRP-152 cells. Thin line = NRP-152; thin line = 152-S3c. Panel D : Transfected BPH-1 cells. Thick line = BPH-1; thin line = BPH-S3c. Panel E: Immunoprecipitation followed by Western blot showing EGFP expression in transfected NRP-152 and BPH cells. Note the lack of EGFP bands for parental lines NRP-152 and BPH-1, whereas EGFP was detected using EGFP-specific Ab (Pharmingen) in lanes for 152-S3c and BPH-S3c. Methods: NRP-152, 152-S3c, BPH-1, and BPH-S3c cells were lysed in buffer. Equal amounts of protein in cell lysates were pre-cleared with Protein A/G beads, then precipitated with anti-FLAG AB plus Protein A/G beads with rotation in the cold. The pelleted beads plus proteins were separated on 12% SDS gels, transferred to PVDF membranes, then blotted with Ab to EGFP. Enhanced chemifluorescence was used to reveal the 27 kD bands corresponding to EGFP. 152-S3c Cells Grew in the Absence of Exogenous Growth Factors To demonstrate that 152-S3c cells grew in the absence of growth factors required by untransfected NRP-152 cells, transfected and untransfected NRP-152 cells were grown in microtiter wells. Proliferation was quantified by the oxidation of MTT after 48 hr. Figure 3 shows the results of these experiments. NRP-152 and 152-pIRES cells grew more slowly in unsupplemented 154 medium than they did in 152 medium. However, 152-S3c cells (3 representative clones, D5, A12, and H4, are shown) grew nearly as well in 154 medium as in 152 medium, and grew significantly better in 154 medium than either NRP-152 or 152-pBABE cells (p < 0.001; Figure 3A ). Therefore, clones of 152-S3c cells, stably transfected with pBABE-S3c, grew in vitro as if they lost the requirement for additional growth factors in the cell culture medium. Figure 3 Growth of NRP-152, NRP-154, 152pBABE, and 152-S3c clones on 154 medium compared to growth on 152 medium. 10 3 cells were seeded in microtiter wells, in the indicated medium. After incubation for 48 hr, MTT (15 μl at 25 μg/ml) were added to each well, and incubation was continued for 4 hr more. The formazan was dissolved in 0.1% SDS, and the absorbance was quantified on a DynaTech plate reader at 570 nm. Unpaired Student t-tests (InStat3 software) were performed to assess the statistical significance of the growth of S3c-transfected cells relative to pBABE transfected and untransfected NRP-152 cells. Panel A: Comparison of growth as measured by MTT absorbance at 48 hours; Panel B: Comparison of growth rates over 72 hours. Stable Expression of S3c in BPH-1 Cells Resulted in STAT3-Dependence for Survival In order to show that the persistent expression of activated STAT3 was required for the survival of the transfected cells, as we have previously shown for hormone-refractory prostate cancer cells lines [ 11 , 12 ], we transfected pIRES-S3c into human BPH-1 cells [ 38 ] for studies with antisense STAT3 oligonucleotides. We used BPH-1 cells and transfected lines only for these experiments, because the antisense oligonucleotide was designed for use in human cells, and we wanted to maximize the efficacy of the antisense oligonucleotide. Figure 4 shows that transfection of 125 nM of sense STAT3 oligonucleotide decreased viability by only 5% at 48 hours, whereas transfection of the same amount of antisense STAT3 oligonucleotide decreased viability to 18% at 48 hours. Furthermore, transfection of antisense STAT3 oligonucleotide into untransfected BPH-1 cells did not decrease viability any more than did transfection of sense oligonucleotide. Figure 4B shows that 24 hours after transfection with 125 nM of antisense STAT3, BPH-S3c cells displayed a 66% reduction in intracellular STAT3 protein levels. We concluded from these experiments that the S3c expressed in BPH-S3c cells was functionally active, and that BPH-S3c cells were dependent upon continued STAT3 expression for their very survival, just like hormone-refractory prostate cancer cell lines [ 11 , 13 ]. These data are more evidence for a profound difference in phenotype between BPH-1 cells and BPH-S3c cells. Figure 4 Functional activity of STAT3 in S3c-transfected cells. Panel A: To show the functional activity of STAT3 expressed by the S3c gene, BPH-1 cells stably transfected with pIRES-S3c were treated with either 125 nM sense or antisense STAT3 oligonucleotide. Percent viability over time was determined by staining with propidium iodide, then quantifying fluorescence on a FACScan flow cytometer. Panel B: Treatment with 125 nM antisense STAT3 oligo reduced the amount of intracellular STAT3 protein in the clone of BPH-S3c cells shown in 3A. Twenty-four hours after transfection, BPH-S3c cells were harvested, fixed, and permeabilized, then stained with antibody to STAT3, as described in Materials and Methods. Quantification was performed on a FACScan flow cytometer. The black line indicates the amount of intracellular STAT3 in BPH-S3c cells treated with sense STAT3, while the grey line shows the amount of STAT3 in BPH-S3c cells given antisense STAT3. STAT3 expression was reduced by 66% in this experiment. 152-cS3 Cells Have Decreased Expression of RAR-β and -γ mRNA, and Increased Expression of RAR-α mRNA In prostate cancer cell lines and archived specimens, we previously found that RAR-β and -γ have decreased mRNA levels, while RAR-α mRNA increased, relative to non-malignant prostate cell lines and the normal margins of the same specimens [ 34 , 35 ]. This finding is also true of NRP-152 and NRP-154 cells: the expression of RAR-β and -γ is decreased in NRP-154 cells relative to NRP-152 cells. In order to see if the same change in retinoic acid receptor subunit expression occurred when S3c is expressed, which is consistent with the malignant phenotype, we did the following experiments. For these, we used 152-S3c and 152-pIRES cells, so that we could compare the RAR levels with those of NRP-154 and parental NRP-152 cells, because these 2 related cell lines are believed to represent two stages in the progression and development of prostate cancer [ 36 , 41 ]. Figure 5 depicts the northern blot hybridization results for RAR-β (Figure 5A ) and -γ (Figure 5B ) in transfected and untransfected cells. Lane 1 in both panels shows the hybridized mRNA for untransfected NRP-152 cells, while both lanes 2 show the hybridized band for NRP-154 cells. Note the decreased amount of RAR-β and -γ in lanes 2 (from NRP-154 cells, the prostatic carcinoma line) relative to the amount in lanes 1, obtained from NRP-152 cells, the benign prostatic hyperplasia line. Lanes 3 show the hybridized mRNA obtained from NRP-152 cells transfected with the vector, pIRES-EGFP, while the bands displayed in both lanes 4 shows that when NRP-152 cells were transfected with pIRES-S3c, the hybridization of RAR-β and -γ decreased similarly to what is observed in lanes 1 and 2. Figure 5C compares RAR-α mRNA expression in the 4 cell lines: lane 1 again is NRP-152 and lane 2 is NRP-154; there is more mRNA hybridized in lane 2 than in lane 1, and the band appears as a doublet in lane 2 as well. Lane 3 shows the results from NRP-152 cells transfected with pIRES-EGFP, while lane 4 shows the results from NRP-152 transfected with pIRES-S3c: note the similar pattern to that of lanes 1 and 2 – lane 4 shows more hybridization and a doublet band for RAR-α as well. We concluded from these results that transfection of NRP-152 cells with pIRES-S3c, but not pIRES-EGFP, induced a change in RAR mRNA expression that is often observed in prostate cancer cell lines and archived specimens. Figure 5 S3c expression inhibited RAR-β and -γ expression and increased expression of RAR-α in NRP-152 cells. Panel A: Effect of S3c on RAR-β mRNA levels. Panel B: Effect of S3c on RAR-γ mRNA levels. Panel C: Effect of S3c on RAR-α mRNA levels. NRP-152, NRP-154, NRP-pBABE, and 152-S3c cells were grown to confluence, and RNA was harvested as described in Materials & Methods. Electrophoretic separation of RNA was followed by transfer to nitrocellulose, then hybridization with 32 P-labeled probe, followed by autoradiography. Lane 1 = NRP-152 (rat benign prostatic hyperplasia line); lane 2 = NRP-154 (rat prostatic carcinoma line); lane 3 = 152-pBABE; lane 4 = 152-S3c. The comparison to 18S RNA is shown for each. BPH-S3c Cells Were Androgen-Insensitive In many human prostate cancers, overexpression of the androgen receptor has been noted [ 42 , 43 ]. Therefore, the development of the hormone-refractory state apparently occurs even when there is no disruption of the expression of the androgen receptor, at least in some prostate cells. To clarify these contradictory data and to check for the development of functional androgen-insensitivity, we examined the growth rate of human BPH-1 and BPH-S3c cells in the presence and absence of dihydrotestosterone (DHT), and also DHT in the presence of the antagonist flutamide (F). Our results, presented in Table 2 , show that while BPH-1 cells respond to DHT and are blocked by F, the same is not true of BPH-S3c. Thus, the persistent expression of S3c in BPH-1 cells resulted in a functionally androgen-insensitive state for these cells. Table 2 Androgen-Insensitivity is conferred by S3C expression in BPH-1 cells Cell nM DHT %Stimulation μM F + nM DHT %Inhibition BPH 10 200 1 10 97 BPH-pIRES 10 250 1 10 99 BPH-S3c 10 2 1 10 -4 DU145 10 3 1 10 3 Cells were grown in 96-well plates for 72 hr, in the presence or absence of drugs (DHT = dihydrotestosterone; F = flutamide, bolded and underlined ) as indicated. The MTT assay was used to assess proliferation. %Enhancement = (absorbance + drug/absorbance - drug) × 100; % Inhibition = 1-(absorbance + drug/absorbance - drug) × 100. The responses of DU145 cells, a human prostate cancer cell line that is androgen-insensitive and resistant to flutamide, is shown for comparison. 152-S3c Cells Lost Sensitivity to the JAK2 Inhibitor AG490 In non-malignant cells, the activation of STAT3 is effected by a specific upstream kinase, JAK1 or JAK2 or sometimes Tyk2. Previously we had shown that the constitutive activation of STAT3 in NRP-154 cells rendered those cells insensitive to apoptosis induced by the JAK2 inhibitor AG490 [ 10 ]. In order to see if insensitivity to AG490 was conferred on 152-S3c cells, we added AG490 to cells and assessed apoptosis 48 hr later by annexin V binding and PI inclusion. Table 3 shows the data we obtained. Whereas NRP-152 and 152-pIRES cells were 45 ± 10% and 38 ± 5% apoptotic, respectively, 48 hr after treatment with 100 μM AG490, only 6.3 ± 3% of 152-S3c cells and 7.5 ± 4% of the NRP-154 cells were apoptotic after 100 μM AG490 treatment. We conclude from these experiments that S3c expression in NRP-152 cells decreased their sensitivity to AG490, which is consistent with what we observed in malignant NRP-154 cells. Table 3 NRP-152 Cells Transfected with S3c lost sensitivity to JAK2 inhibitor AG490 Cell S3c? Rx μM % Apoptotic +/- SD NRP-154 YES AG490 0 8 ± 4 100 7.5 ± 5 5152-S3c YES AG490 0 7.5 ± 4 100 6.3 ± 3 152-pIRES NO AG490 0 11 ± 2 100 38 ± 5* NRP-152 NO AG490 0 7.5 ± 4 100 45 ± 10* Cells were placed in 60 mm wells for 48 hr with compound at the concentrations indicated. Zero concentration of the compound is the vehicle (DMSO) control. At the end of the incubation period, cells were harvested, washed, and stained with FITC-annexin V, to demonstrate apoptotic cells. Quantification of fluorescence was performed on a Becton-Dickinson FACScan flow cytometer using CellQuest software. * p < 0.005 by Student t-test, compared to vehicle-treated cells. 152-S3c Cells Grew in Soft Agar As an in vitro indication of tumorigenic potential, soft agar cloning assays were performed as described [ 44 ]. S3c-transfected cells were compared to NRP-152 and to pIRES-EGFP-transfected cells in these experiments. We observed that 152-S3c cells grew significantly better (p < 0.0001 by 2-tailed Student t-test) in soft agar than either untransfected NRP-152 or pIRES-transfected NRP-152 cells (Table 4 ). We conclude from these experiments that 152-S3c cells have the potential to form tumors in vivo, whereas it has previously been established that NRP-152 cells are not tumorigenic [ 36 ], and we would not expect 152-pIRES cells to be tumorigenic either. Table 4 NRP-152 Cells transfected with S3c grew in soft Agar CELL S3c? #WELLS #COLONIES +/- SEM NRP-152 NO 12 2.6 ± 0.9 152-pIRES NO 12 5.8 ± 1.8 152-S3c YES 12 35 ± 4.5* Cells were placed in 60 mm wells in soft agar for 10 days. Colonies of more than 1 mm in size were counted were counted using an inverted phase contrast microscope. * p < 0.001 by Student t-test Expression of S3c Did Not Confer Tumorigenicity on Benign NRP-152 Cells Based on our previous data, especially the soft agar cloning data, we expected that 152-S3c cells would form tumors in SCID mice. However, in 3/3 experiments (in two them, Matrigel was used to enhance tumorigenicity of the cells), an average of 1/5 mice developed tumors; these were 1 mm in diameter or less. We chose to use only transfected NRP-152 cells for these experiments, because in certain in vivo environments, untransfected BPH-1 cells have been observed to form tumors [ 38 ]. We conclude that while persistent S3c expression altered the phenotype of 2 different benign prostatic hyperplasia lines in ways consistent with the development of the malignant phenotype, an additional change in gene expression may be required for tumorigenicity in prostate cancer development. Discussion We have demonstrated that NRP-152 and BPH-1 cells transfected with a constitutively-activated form of the STAT3 gene, S3c, gained a phenotype which more closely resembled that of NRP-154 cells. Specifically, the transfected cells expressed resistance to the antibiotic G418, and also expressed the FLAG epitope, as revealed by intracellular flow cytometry following staining with anti-FLAG Ab in Figure 2B–C , while Figure 2A shows the FLAG expression in mock transfected cells. As additional evidence of S3c expression, we looked for EGFP expression in 152-pIRES cells, since the bicistronic message from this vector (pIRES-EGFP) places the S3c gene 3' to the EGFP, so that S3c would have to be translated before EGFP is translated. Figure 2D shows the EGFP expression in the same clone whose FLAG expression is shown in Figure 2C . These results were confirmed by immunoprecipitation/Western blot analysis, which is shown in Figure 2E , whereupon cell lysates were precipitated with Ab to the FLAG peptide on the S3c gene, then blotted with anti-EGFP Ab. Only the transfected and selected 152-S3c and BPH-S3c cells revealed EGFP bands, not the parental lines. After obtaining these results, we characterized the phenotype of the transfected cells. Parental NRP-152 cells are fastidious in their growth factor requirement, whereas NRP-154 cells and 152-S3c clones grew in medium supplemented only with serum (Figure 3 ). Therefore, we assessed the change in growth of transfected NRP-152 cells by comparing their growth in unsupplemented medium. We found that clones of 152-S3c cells grew nearly as well as NRP-154 cells in simple medium, whereas NRP-152 and 152-pIRES cells grew poorly in the absence of growth factors included in the medium (Figure 3 ). The change in growth factor requirement is one often observed for neoplastic cells, and is consistent with the role of STAT3 as a proto-oncogene with the capability of transforming benign cells into malignant cells [ 15 , 45 ]. As for dependence on survival of constitutively-activated STAT3, which has been observed in NIH-3T3 transfected with S3c [ 40 ] and in hormone-refractory prostate cancer cell lines [ 11 ], BPH-S3c cells treated with 125 nM antisense STAT3 oligonucleotides died over time, going from 100% viable to less than 20% viable 48 hours after transfection (Figure 4A ); the reduction in viability could be attributed to the effect of antisense STAT3 on STAT3 protein expression, which was reduced by 66% at 24 hours after transfection (Figure 4B ). These data mean that like hormone-refractory prostate cancer cells, BPH-1 cells transfected with S3c became dependent upon the continued expression of S3c for their survival. As for RAR expression, we observed decreased mRNA levels of RAR-β and -γ, but increased RAR-α expression in S3c-transfected NRP-152 cells; the results shown in Figure 5 are consistent with the expression levels of these receptor mRNAs previously observed by us in prostate cancer lines [ 34 ] and in prostate cancer patient specimens [ 35 ]. These findings are echoed in those of Yang, et al., who observed that IL-6-induced STAT3 signaling in lung epithelial cell lines lead to increased RARα expression, which was abrogated when the STAT3 DNA-binding domain was substituted by the corresponding STAT1 domain [ 46 ]. The importance of our results with respect to prostate cancer is that this disease is often refractory to retinoid therapy, the molecular basis for which is not known at this time. Our results gives possible insight into the mechanism of retinoid insensitivity, and might also indicate that treatment of prostate cancer with STAT3 inhibitors and with retinoids may be beneficial. In terms of androgen receptor function, S3c expression in BPH cells changed their response to androgens so that BPH-S3c cells were no longer stimulated by DHT, and the growth of BPH-S3c cells was not inhibited by flutamide treatment (Table 2 ). These findings with respect to the androgen receptor and responses to DHT and flutamide are especially important, as it may be the one of the first indications of a direct effect of STAT3 on androgen receptor responses, and may indicate a possible molecular mechanism for the development of the hormone-refractory state in prostate cancer patients. The progression to androgen-independence has been found to be associated with IL-6, with c-myc expression, and with insulin-like growth factors, all of which can signal through the activation of STAT3 [ 25 - 28 ]. It has been postulated that cross-talk between STAT3 and the androgen receptor plays a role in the development and maintenance of the hormone-refractory state in prostate cancer [ 47 ]; our data indicate that persistently-activated STAT3 may obviate the need for expression of the androgen receptor, since the androgen receptor did not respond to either DHT or F in S3c-transfected BPH-1 cells (Table 2 ). Further work is warranted in this area. Prior to performing in vivo tumorigenicity experiments, we wanted to see if S3c-transfected cells could grow in soft agar as clones. We observed that S3c expression in NRP-152 cells allowed them to grow as clones in soft agar (Table 4 ). However, even though 152-S3c cells grew in soft agar, a phenotype usually consistent with tumorigenicity, in 3 out of 3 experiments we failed to observe tumors in more than 20% of the mice, and these tumors were not more than 1 mm in diameter (data not shown). Therefore, we concluded from these data that persistent expression of activated STAT3 alone was not sufficient to produce tumorigenicity in prostatic epithelial cells, although it had been sufficient in NIH-3T3 cells, as previously reported [ 40 ]. Furthermore, recent observations by Zhang and coworkers point to an important function for STAT3 in both tumorigenesis and metastasis formation in leiomyosarcoma [ 48 ], due to signaling by hepatocyte growth factor/scatter factor. Among the candidate genes regulated by STAT3 in this regard are matrix metalloproteinase-2, which is essential for tumor invasion and metastasis formation [ 49 ]. Perhaps STAT3 cooperates with another factor regulated by hepatocyte growth factor/scatter factor, which is not expressed by either NRP-152 or BPH-1 cells. Only more experiments will reveal whether this is the case. Indeed, we are planning experiments to see what genes are regulated by S3c, to gain insight into the phenotypic changes induced by S3c expression. For example, very recently it was reported that STAT3 and the microphthalmia-associated transcription factor were both required for optimal upregulation of c-fos , and subsequent tumorigenicity, in NIH-3T3 cells [ 50 ]. Whether the prostatic lines NRP-152 or BPH-1 express microphthalmia-associated transcription factor has not been determined; the levels of c-fos in S3c-transfected lines can be determined. As well, Dechow and coworkers reported that transfection of S3c into mammary epithelial cells rendered those cells tumorigenic in irradiated SCID mice [ 51 ]; whether our results are an indication of a difference between mammary epithelial cellls and prostatic epithelial cells or a reflection of irradiated vs. non-irradiated SCID mice remains to be elucidated. As more information is revealed about gene expression changes that accompany the progression of prostate cancer from the benign to the hormone-refractory state, the other genetic changes needed for tumorigenicity of S3c cells should be revealed. Conclusions Our data indicate that transfection of NRP-152 and BPH-1 prostatic epithelial cells with a gene for persistently-activated STAT3, S3c, changed the phenotype of the cells into one resembling a malignant phenotype, thereby giving even more importance to the role of activated STAT3 in the transformation of normal cells into neoplastic cells. Importantly, we found that cells expressing S3c depended on its continued expression for survival. Two kinds of evidence are presented: first, S3c-transfected cells became sensitive to the effect of antisense STAT3 oligonucleotide. When transfected with antisense STAT3, both BPH-S3c and 152-S3c underwent apoptosis. Second, the S3c-transfected cells were not sensitive to the commonly-used STAT3 inhibitors, which are really JAK inhibitors, because activation of STAT3 by the upstream JAK is not required when S3c is expressed. We observed that growth factor-dependent NRP-152 cells grew without growth factor supplementation when transfected with S3c gene, whereas the medium for vector-transfected NRP-152 cells still required supplementation with growth factors. Moreover, we observed that 152-S3c cells grew in soft agar, whereas neither vector-transfected nor untransfected NRP-152 cells did. Furthermore, we observed that the expression of RAR subunits in 152-S3c cells was different from vector-transfected and untransfected NRP-152 cells, and that the changes were consistent with what we previously observed in specimens from prostate cancer patients, as well as in primary prostatic epithelial cells compared with prostate cancer cell lines [ 34 , 35 ]. These data may have implications for the relative lack of sensitivity of PCA to retinoid therapy. As for BPH-1 cells, which do not require growth factor supplementation, we observed that when transfected with S3c, this cell line lost its responses to testosterone and to the testosterone antagonist flutamide. Neither of these changes was observed in vector-transfected BPH-1 cells. However, neither S3c-transfected cell line was tumorigenic when injected into SCID mice, leading us to conclude that additional genetic changes are possibly needed for tumorigenicity in prostate cells. Methods Cell Lines NRP-152 and NRP-154 cells were the gift of Dr. David Danielpour, Ireland Cancer Center, University Hospitals, Cleveland, OH [ 36 ]. Growth factor-dependent NRP-152 cells were grown in DMEM/Ham's F12 medium (1:1; GIBCO) supplemented with 10% newborn bovine serum (Hyclone), 2 mM glutamine (GIBCO), epidermal growth factor (20 ng/ml), insulin (5 μg/ml), dexamethasone (0.1 μM) and cholera toxin (10 μg/ml; all, Sigma), pH 7.3 (152 medium). NRP-154 cells were grown in DMEM/Ham's F12 medium plus 10% newborn calf serum (154 medium). Growth factor-independent BPH-1 cells [ 37 , 38 ] were the gift of Dr. Simon Hayward, Vanderbilt University, Nashville, TN. They were grown in RPMI-1640 medium supplemented with 10% newborn bovine serum. For transfections, cell were seeded into wells of 6-well plates and grown until 50–80% confluent monolayers of cells were present, as assessed by observation under inverted phase-contrast microscopy. Transfections Derivation of the pBABE-S3c plasmid containing a constitutively-activated STAT3 gene, S3c (gift of Dr. Jacqueline Bromberg, Memorial Sloan-Kettering Cancer Institute) has been previously described [ 14 , 45 ]. The S3c gene was excised along with its FLAG tag, and inserted into pIRES-EGFP (Clontech), resulting in the plasmid called pIRES-S3c. For stable transfections, Clonfectin reagent (Clontech) was mixed with plasmid DNA (6 μl Clonfectin and between 1 and 3 μg plasmid), according to the manufacturer's instructions. The complete medium was removed from the plates of cells and replaced with 1.8 ml IMDM (Invitrogen). The Clonfectin-plasmid mixtures (100 μl) were added to the cells; replicate cultures of cells received Clonfectin only at the time of transfections. The plasmid-Clonfectin mixtures were left on the cells in the incubator for 4 hr, at which time the supernatant fluids were aspirated and replaced with 5 ml/well pre-warmed complete medium. Twenty-four hr following transfections, G418 (Invitrogen) was added at a final concentration of 800 μg/ml. The medium plus G418 was replaced 3 times/wk until no surviving cells were observed on the Clonfectin-only wells, usually 2 weeks. At that time, G418 was added at 100 μg/ml to maintain the transfected cells. When the transfected cells reached confluence, they were used for further analyses. Table 1 gives a summary of transfected cells and phenotypes obtained. For transient transfections, LipoFectamine 2000 in Opti-MEM I medium (both, Invitrogen) was used according to the manufacturer's directions. For subconfluent (~50%) cells, 2 μl of LipoFectamine 2000 was used with varying amounts of antisense or sense STAT3 oligonucleotide (gift of Dr. James Karras, ISIS Pharmaceuticals). The oligonucleotides were left on the cells for 6 hours before cell culture medium supplemented with 30% was added to each well. Cells were incubated until assays were performed. Limit-Dilution Cloning In order to analyze clonal populations of cells, transfected cells (pIRES-S3c or pIRES-EGFP) were harvested, diluted to 10 cells/ml in complete medium, and seeded into microtiter plates at 100 μl/well. The total volume of each well was brought to 200 μl with additional medium, and the plates were incubated until growth of seeded cells was observed, usually at 10 days to 2 weeks. Determination of Stable Transfection by Expression of FLAG in 152-S3c and BPH-S3c Cells by Intracellular Flow Cytometry Expression of the FLAG epitope engineered onto the constitutively-activated STAT3 gene in transfected NRP-152 cells was performed by intracellular flow cytometry, as described [ 52 ]. Briefly, 152-S3c or BPH-S3c cells were harvested, washed, and fixed in 4% paraformaldehyde/PBS (Pharmingen) for 30 min on ice. Fixed cells were washed and permeabilized with 0.1% sapononin (Pharmingen) for 15 min at room temperature, then washed. Mouse monoclonal Ab M1 to FLAG (Sigma) was added (1 μg/10 6 cells/100 μl permeabilization buffer) to the cells for 1 hr on ice. The cells were washed 3 times, then incubated with phycoerythrin (PE)-labeled goat anti-mouse F(ab 2 )' (Caltag) for 1 hr on ice in permeabilization buffer. After washing 3 times, cells were resuspended in 1 ml PBS and analyzed on a Becton-Dickinson FACScan. CellQuest software was used to acquire and analyze the fluorescence. The Kolmogorov-Smirnov 2-sample test was used to determine the level of significance of the change in fluorescence intensity between control-stained (F(ab 2 )'-stained only) and Ab-stained populations of cells, thereby ascertaining that the populations observed in the histograms were truly separate populations of cells [ 53 ]. Immunoprecipitation/Western Blot Studies For immunoprecipitation, cells lysed in Lysis Buffer (10 mM PBS, pH 7.4, 1% NP-40, 0.5% sodium deoxycholate, 0.1% sodium dodecylsulfate (SDS), 1 mM sodium orthovanadate, 1 mM phenylmethyl-sulfonyl fluoride, 40 μg/ml aprotinin) were precleared with Protein A/G agarose (Santa Cruz Biotechnology), then precipitated with 1–5 μg rabbit Ab (Cell Signaling or Pharmingen) plus Protein A/G (Santa Cruz Biotechnology) agarose overnight. After washing, the beads were eluted by heating in Laemmli buffer for 5 min at 95°C, followed by electrophoretic separation on 12% SDS-polyacrylamide gels (Novex Nu-PAGE pre-cast gels). Transfer of separated protein species to nylon membrane (Millipore) was followed by blocking in 10% non-fat dry milk in TBST (50 mM Tris HCl, pH 7.4, 150 mM NaCl, 0.3% Tween 20). Incubation of the membrane with rabbit Ab was followed by incubation with alkaline phosphatase-linked goat anti-rabbit antibody (Amersham ECF kit). After addition of substrate from the kit, the membranes were read by the Typhoon imager, with ImageQuant software for resolution of images (Molecular Dynamics). Measurement of In Vitro Growth of Cells NRP-152, NRP-154, BPH-1, and transfected cells were seeded at 10 3 cells/well in microtiter plates in appropriate medium, as indicated. After 48 hr, 15 μl MTT (Sigma; 25 μg/ml) was added to each well for 4 hr, then the resulting formazan was dissolved in 0.1% SDS. Absorbance was determined at 570 nm on a Dynatech microplate reader. Statistical determinations of significance were performed by unpaired Student t-test for multiple independent assays, using GraphPad software. Determinations of Androgen Insensitivity and Presence of Retinoid Receptors The effect of dihydrotestosterone (DHT) as growth agonist, and the effect of flutamide (F) as growth antagonist, was assessed by use of the MTT assay described above. DHT and F were obtained from Boeringer-Mannheim, and cells were treated with 1 or both drugs at concentrations ranging from 1 to 100 nM for DHT, and 0.1 to 3 μM for F. These are within the published ranges of efficacy for these drugs [ 34 , 54 ]. Vehicle controls were included. Replicate plates were harvested at 24, 48, 72, and 96 hrs after treatment. Northern blot hybridizations to detect the retinoid receptors RARα, RARβ, and RARγ were performed as previously published [ 34 ]. In brief, RNA was isolated from cells using RNAEasy (Qiagen) and quantified spectrophotometrically. RNA was separated by size on agarose gels, then transferred to nitrocellulose membranes (Schleucher & Schuell). The probe was labeled with 32 P-dCTP (New England Nuclear), then allowed to hybridize to the blot overnight in hybridization buffer. After washing, hybridization was detected by use of a PhosphoImager (Molecular Dynamics). Apoptosis Assays Forty-eight hr after transient transfection, cells were harvested using Enzyme-Free Cell Dissociation Buffer. After two washes with PBS, they were stained with FITC-annexin V (5 μl/10 6 cells; Caltag) for 15 min at room temperature. Apoptotic cells (cells staining with FITC-annexin V) were quantified by measuring green fluorescence in FL1 on the flow cytometer. In some experiments, cells were also stained with propidium iodide (PI), which is detected by the FL3 detector. CellQuest software was used to acquire and analyze the data on a Becton-Dickinson FACScan flow cytometer. For studies using the tyrphostin JAK2 inhibitor AG490 (Calbiochem), the dissolved compound was added to subconfluent cells, as described [ 11 ]. A vehicle control was included for the 0 μM concentration. Forty-eight hrs later, cells were harvested and processed for quantification of apoptosis by annexin V binding and PI incorporation. Assay for Growth in Soft Agar Transfected cells were subjected for growth in soft agar to assess their change in phenotype with regards to colony formation. After selection and cloning, 10 4 cells were trypsinized and washed in Ca2+/Mg2+-free PBS (Life Technologies) and plated in 1 ml of medium plus serum without supplements containing 0.3% (w/v) Noble Agar (Difco/Becton-Dickinson) over a 2 ml layer of the same medium with 0.6% agar in six-well plates. The number of colonies was counted using low magnification microscope (4×) after 10 days. In Vivo Tumorigenicity Studies Our protocol was reviewed and approved by the Institutional Animal Care and Use Committee of UMDNJ. Severe-combined immunodeficient (SCID) mice (Charles River Laboratories) were obtained at 5 weeks of age, and acclimatized in the barrier vivarium for 1 week. At that time they were injected subcutaneously with 8 × 10 6 S3c or vector-transfected (pIRES-EGFP) control cells. Each group consisted of 5 animals. In some experiments, the cells were mixed with Matrigel (Collaborative Research) prior to injection. Tumor growth was monitored weekly using engineer's caliper's to measure the 2 perpendicular diameters, over the course of 12 weeks. List of abbreviations STAT signal transducer and activator of transcription cSTAT3 constitutively-activated STAT3 JAK Janus activated kinase PCA prostate cancer S3c constitutively-activated STAT3 gene having a Cys substitution FLAG an immunogenic peptide fused to gene of protein to be expressed for identification and/or purification purposes SCID severe combined immunodeficient PBS phosphate-buffered saline DMEM Dulbecco's modifcation of Eagle's medium IMDM Iscove's modification of Eagle's medium FITC fluorescein isothiocyanate PE phycoerythrin PI propidium iodide FL1, 2, or 3 fluorescence detectors on a flow cytometer that collect fluorescence data within a set range of wavelengths, FL1 being the lowest and FL3 being the highest EGFP enhanced green fluorescence protein RAR retinoic acid receptor DHT dihydrotestosterone F flutamide Authors' contributions HFH conceived of the retinoic acid receptor subunits experiments, and perofrmed the northern blot hybridizations. TFM performed the growth factor dependence and growth rate experments, performed some of the in vivo experiments, and prepared cells for flow cytometry. PS performed some of the transfections, participated in the in vivo experiments, the western blots, and prepared cells for flow cytometry. ABB made the pIRES-S3c plasmid from pBABE-S3c, and performed the soft-agar cloning experiments. BEB conceived of the project, performed most of the transfections, performed some of the in vivo experiments, and performed all flow cytometry acquisitions and analyses. All authors read and approved the manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC546221.xml
544191	A suite of web applications to streamline the interdisciplinary collaboration in secondary data analyses	Background We describe a system of web applications designed to streamline the interdisciplinary collaboration in outcomes research. Description The outcomes research process can be described as a set of three interrelated phases: design and selection of data sources, analysis, and output. Each of these phases has inherent challenges that can be addressed by a group of five web applications developed by our group. QuestForm allows for the formulation of relevant and well-structured outcomes research questions; Research Manager facilitates the project management and electronic file exchange among researchers; Analysis Charts facilitate the communication of complex statistical techniques to clinicians with varying previous levels of statistical knowledge; Literature Matrices improve the efficiency of literature reviews. An outcomes research question is used to illustrate the use of the system. Conclusions The system presents an alternative to streamline the interdisciplinary collaboration of clinicians, statisticians, programmers, and graduate students.	Background In the last decade, the number of relevant data sources available for outcomes research has grown exponentially. In contrast, the number of individual researchers with clinical and statistical expertise required to explore these data sets increase at a much slower pace. As a result, an immense quantity of valuable clinical data are left untouched, never becoming clinical publications that could potentially improve health care. The disproportion between data volume and number of qualified researchers can be explained by the growing complexity involved in outcomes research projects using secondary data analyses. Researchers have to formulate of a clinically relevant and methodologically sound research question, find appropriate data sources, perform statistical analyses, and generate a final manuscript that will be submitted for peer-review. Frequently, individual researchers have the training and time to perform a few of these steps, but the integration of all tasks calls for an interdisciplinary systems approach [ 1 , 2 ]. This interdisciplinary effort, however, is often challenged by communication problems among researchers with different backgrounds, particularly when physicians with an exclusive clinical education attempt to work in collaboration with quantitative researchers such as statisticians [ 3 ]. As a consequence, the output of such collaboration is either scarce or absent. This article describes a suite of web applications developed to facilitate the process of converting outcome databases into clinical manuscripts, to streamline the interdisciplinary collaboration of researchers, and to connect all different steps of the outcomes research process. To illustrate its use, we will describe how a research project has been conducted using this system from its early phase of research question formulation to the completion of the final manuscript. Construction and content The system of Web applications is composed by five different tools: QuestForm, Research Manager, Analysis Charts, and Literature Matrices. These tools were designed to assist researchers in each of the phases encountered in an outcomes research project involving secondary data analysis (Figure 1 ). All tools are freely available at a designated web site . The following sections will describe each of the Web applications and their application in the answer of a real outcomes research question. Figure 1 Research phases, challenges, and respective tools QuestForm General description QuestForm, an acronym for "Question Formulation", is an application designed to assist researchers in the location of clinical databases and formulation of outcome research questions (Figure 2 ). Clinical databases contain raw data (observations from individual patients) from national administrative claim data, cohort studies, clinical trials, and registries (see for an updated list). All databases have been de-identified and do not contain protected health information as specified by the Health Insurance Portability and Accountability Act ( , accessed on Aug/04/2004). The application The application is built using Extensible HyperText Markup Language 1.0 (XHTML) [ 4 ], Java, and a relational database (MySQL 4.0)[ 5 ]. Figure 2 QuestForm application – Database search engine QuestForm starts by presenting researchers with three main strategies to find research databases: use of pre-determined key words that describe the database as a unit (Figure 2 ), user-defined key words to describe variables present in the data dictionary of each database, and the presentation of a complete list of all databases. Once databases are located, researchers can read an overall summary about the database including details about number of subjects, sampling strategy, data ownership, and overall characteristics of the study population and associated procedures (Figure 3 ). Researchers then determine that the database most appropriate to answer the research question at hand, a JAVA screen is displayed for research question formulation (Figure 4 ). This screen presents all variables displayed in hierarchical categories. Variables are presented with the corresponding question and alternative responses. All variables can be inserted into a research question (Question Diagram) divided into the classical categories for an epidemiological question: Outcomes, Predictors, Confounders, Inclusion and Exclusion Criteria. Search engines are provided for ICD9-CM diagnosis and procedure codes (Figure 5 ), which can also be inserted into the Question Diagram. Finally, previously formulated Question Diagrams can be shared among researchers. This latter functionality allows researchers to both share Question Diagrams among members of the ongoing project as well as share previously formulated Question Diagrams with researchers from other teams. Once the question is fully formulated, researchers can save the question as in a graphical format known as Question Diagram. Figure 3 QuestForm application – Overall description of the database Figure 4 QuestForm application – Formulation of a Question Diagram Figure 5 QuestForm application – Search engine for ICD9 codes Outcomes research application Dr. Guller initiated the project searching for an existing database that would allow him to compare surgical outcomes between laparoscopic and open appendectomy procedures in the treatment of acute appendicitis. The outcomes were pre-specified as mortality and infection, although other existing outcomes would also be of interest. Although there are multiple single-institution studies attempting to answer this question [ 6 , 7 ], few studies have taken a population approach to test whether one procedure is superior to the other. As a first step, Dr. Guller searched across previously formulated Question Diagrams to evaluate whether other studies could have used a similar research design. Since none was found in QuestForm, Dr. Guller searched across more than forty different databases for an existing database that would have the variables to answer his research question. After navigating through multiple data dictionaries, Dr. Guller found that the Nationwide Inpatient Sample (NIS) Release 6, 1997 [ 8 ] presented the variables and an adequate number of patients to answer his question. Once the database was located, Dr. Guller selected the outcomes of interest (length of hospital stay, in-hospital complications, in-hospital mortality, rate of routine discharge), main predictors (laparoscopic versus open procedures), and confounders (age, gender, race, household income, comorbidity, hospital volume, location of the hospital, teaching status of hospital, and appendix perforation), inserting each of them into the research question fields in QuestForm. Using built-in search engines for ICD9 codes, Dr. Guller created the definition for each of the above-mentioned variables and defined the inclusion and exclusion criteria. The final research question was then saved as a Question Diagram and immediately submitted to Dr. Pietrobon for feasibility evaluation. Dr. Pietrobon judged that the project was feasible and could be completed using the database indicated by Dr. Guller. At this point, the project was initiated and a detailed project management plan was established using Research Manager. Research manager Research Manager is a Web application developed by our group designed to facilitate the project management of clinical research projects. Similar to QuestForm, Research Manager is licensed under the GNU Public License [ 10 ], which allows individuals to copy, modify, and freely distribute the software as long as the source code is provided. Research Manager provides multiple features to facilitate project management of clinical research projects. All projects are displayed by category (e.g., cardiology, general surgery, etc) with a brief description. The internal content of all projects is password protected. All internal tasks within a project are assigned to individual researchers. Project administrators initially assign deadlines that can be modified by task leaders within three days. All participating members of the project receive weekly reports containing details about the activity and the latest electronic file within each task (Figure 6 ). These files can include research questions, data analysis files, synthesis of a literature review, and manuscript drafts. Project members can also customize the application to receive updates for every single file uploaded to Research Manager in real time if they decide to closely track the project. Expired tasks are marked in the weekly report sent to the entire team, thus providing an incentive for investigators to keep tasks within planned deadlines. Figure 6 Research Manager Research Manager helps identifying such problems and enables their early elimination, thus avoiding delays in the project completion. Finally, since weekly reports are generated to all participating members, Research Manager also provides peer-incentive for project members to complete their tasks in a timely manner. Outcomes research application Once the Question Diagram was evaluated by the clinical epidemiologist, this project was transferred to Research Manager. Contact information for each of the researchers involved and deadlines for completion of the main phases of the research process were set, including data extraction and cleaning, data analysis, literature review, and manuscript writing. Weekly reports were generated to update investigators on all tasks of the project, including different versions of the statistical analysis, modifications in the research question, and manuscript sections. With an established project plan, the statistician in charge selected the best methods for analysis. Since the database is a random sample of the United States and requires special survey analysis methods, it was necessary that all involved researchers understood the statistical approach by using Analysis Charts. Analysis charts Analysis Chart is a tool designed to enhance the understanding of statistical methods to a format that is understandable by clinical researchers with different previous levels of statistical knowledge. As such, it is important in the design as well as the analysis phases of a project (Figure 7 ). The application was built using Extensible HyperText Markup Language (XHTML 1.0) [ 4 ] in combination with Cascading Style Sheets [ 11 ]. Figure 7 Analysis Chart Analysis charts are composed by cascading links that display information about quantitative methods in progressive levels of complexity called "layers of information". Each layer explains the statistical method with an increasing level of complexity. In this manner, clinicians interested in simply understanding the method to evaluate whether it can be applied to a research project can simply read the first three layers. In contrast, researchers interested in a direct application of the method to available research data can follow all layers and their respective references. Most commonly, complex techniques are presented using five layers of information . Layer 1 summarizes the general goal of the method. Layer 2 presents previous clinical applications so that the researcher can visualize situations in which the method may be realistically applied. Layer 3 describes the data requirements for the application of the statistical technique. Layer 4 describes the basic statistical underpinnings of the method, initially breaking down equations and then reassembling them. Finally, layer 5 presents a list of available software packages for the implementation of the technique as well as cases studies where all previous layers are applied to real data sets. Each layer ends with a section containing selected references that explain the topic in more detail. Outcomes research application While deciding on the most appropriate analysis strategy for the Question Diagram, Drs. Pietrobon and Guller consulted the Analysis Chart searching for the most appropriate statistical methods of analysis. Given the nature of the research question and that the NIS database has a sample design, Drs. Guller and Pietrobon opted for an approach involving multiple and logistic regression models while adjusting for sampling weights, strata, and clusters. With a defined analysis protocol, the research question was then transferred to a statistical programmer trained in the translation of Question Diagrams into statistical code. This process was closely evaluated by Drs. Pietrobon and Guller, who scrutinized the statistical code and results from a clinical and statistical perspectives. Once the results were deemed to be accurate, Dr. Pietrobon started the literature review using a Literature Matrix. Literature matrices Literature matrices consist of a comprehensive but not necessarily exhaustive review of the literature focused on a narrow clinical topic (Figure 8 ). Each article is analyzed using the following criteria: study objectives, data sources, outcome variables, primary predictor variables, confounders, statistical analysis, results, established knowledge, and shortcomings. Each literature matrix is saved as an XHTML file that can be visualized in web browsers as well as imported in any commercial or open source spreadsheet applications such as Microsoft Excel ® or Open Office Calc [ 12 ]. Figure 8 Literature Matrix The advantage of the Literature Matrix as implemented in the Web suite is its availability over the web to the whole outcomes research community. This allows Literature Matrices to be constantly updated, with authors receiving their due credit in a list of contributors. Literature Matrices also enable researchers to obtain a complete summary of the literature without going through the cumbersome process of copying and reading a manuscript for the first time. In contrast, researchers' time can be spent more efficiently in reviewing what has already been compiled and attempting to expand the Literature Matrix with other relevant bibliographic references. Outcomes research application In order to evaluate the literature, a graduate student performed a thorough literature search. All relevant articles were copied, read, and the data extracted according to the established categories. Once the Literature Matrix had been completed, Drs. Guller compared the current project results with results published in the literature. The structured information in Literature Matrices also allowed Dr. Guller to compare the strengths and weaknesses of the current project in relation to previous publications. Once this phase was completed, Dr. Guller proceeded to the final writing of the manuscript using Output Templates. Outcomes research application At the end of the project, Dr. Guller combined the Question Diagram, Analysis Chart, and Literature Matrices to write the final manuscript. While Analysis Charts provided the information concerning the statistical techniques used in this study, Literature Matrices provided the basis for comparison of the study results against previous publications. Although not included in the final manuscript, Dr. Guller also had access to multiple analysis files through Research Manager to orient him in each of the steps taken during the research question formulation, data analysis, and literature review. Use of web application by clinical researchers The use of web applications by individual clinical researchers can be summarized in Figure 9 . Figure 9 Integration between UR tools and clinical researchers Utility Since the concept of streamlining the interdisciplinary collaboration preceded the existence of the current suite of web applications, different versions of the central idea have been gradually applied since the second half of 2002. Although our usability, qualitative, and economic studies to evaluate this application are still ongoing, we have noticed a significant improvement in the number and quality of our publications as evidenced by the increasing acceptance rates of our manuscripts for publication in peer-reviewed journals. The Web applications are currently used in research projects involving Duke University and other universities in the United States and abroad, where they have been shown to facilitate inter-institutional collaborations. Discussion Improving the efficiency of an interdisciplinary approach to secondary data analyses has multiple potential benefits. These include an increase in the overall clinical significance of the final publication, decrease in the number of failed projects, decrease in the time for completion of individual projects, improvement in the education of future outcomes researchers, decrease in the cost-benefit ratio for individual outcomes research projects, and, perhaps most importantly, an automation of repetitive tasks. This last factor is crucial since eliminating repetitive tasks will allow researchers to concentrate on the design of innovative projects [ 2 ]. Surprisingly, few systems and applications have been described to solve the problem of complex interdisciplinary collaboration between clinicians and statisticians. Isolated approaches have usually focused on specific portions of the outcomes research process, without attempting to integrate them into a cohesive system. For example, Marshall [ 13 ] has proposed the use of a secure Internet web site for collaborative medical research and data collection. While this system seems to achieve its proposed objectives, it does not improve the process of guiding teams in the translation of data into useful clinical information. Other systems have approached the process in a more comprehensive manner. The Research Toolbox [ 14 ], for example, is a software application that combines databases for literature searches in addition to providing templates for the scientific output. The system is applicable to any type of research, but lacks the ability to connect researchers over the World Wide Web. It also does not address the formulation of research questions from existing databases, selection of statistical techniques, exchange of manuscripts, or project management. Finally, the Web-based Medical Information Retrieval System (WebMIRS) project, funded by the National Library of Medicine [ 15 ], allows researchers not only to evaluate the database content but also to perform the data extraction of specific subsets of the data set. In spite of its high performance as a research tool, WebMIRS is currently restricted to one single publicly available database (National Health and Examination Survey – NHANES), and does not contribute to other phases of the outcomes research process. Although this newly developed system of applications provides a significant improvement in the way secondary data analyses are conducted, it still has limitations. First, because of the lack of a formal evaluation of the effectiveness of this system, we are unable to quantify its real time and cost saving benefits. Second, the system is currently restricted to secondary analyses and does not allow for the planning of prospective data collection. Although one of the main advantages of formulating a research question based on existing data sets is the bounded nature of the process, future applications should attempt to create rules and algorithms that may guide prospective data collection. Conclusion In summary, we have experienced that this system has significant advantages over the traditional manner of conducting outcomes research based on secondary data analyses. This tool may have important applications, not only resulting in an improvement in the overall efficiency of the outcomes research process, but also affecting the way new outcomes researchers are trained and introduced to a research environment. Availability and requirements The Web application is available at Abbreviations XHTML Extensible HyperText Markup Language 1.0 JAVA: by Sun Microsystems ICD9-CM: International Classification of Diseases, Ninth Revision – Clinical Modification NIS: Nationwide Inpatient Sample GNU: GNU's Not linux General Public License WebMIRS: Web-based Medical Information Retrieval System NHANES: National Health and Examination Survey UR: Uniform Resource Competing interests The author(s) declare that they have no competing interests Authors' contributions Ricardo Pietrobon – design, manuscript drafting Ulrich Guller – design, manuscript revision for important intellectual content Henrique Martins – design, manuscript revision for important intellectual content, software programming Andreia P Menezes – design, manuscript revision for important intellectual content Laurence D. Higgins – design, manuscript revision for important intellectual content Danny O. Jacobs – design, manuscript revision for important intellectual content Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC544191.xml
545072	Characterisation of cytotoxicity and DNA damage induced by the topoisomerase II-directed bisdioxopiperazine anti-cancer agent ICRF-187 (dexrazoxane) in yeast and mammalian cells	Background Bisdioxopiperazine anti-cancer agents are inhibitors of eukaryotic DNA topoisomerase II, sequestering this protein as a non-covalent protein clamp on DNA. It has been suggested that such complexes on DNA represents a novel form of DNA damage to cells. In this report, we characterise the cytotoxicity and DNA damage induced by the bisdioxopiperazine ICRF-187 by a combination of genetic and molecular approaches. In addition, the well-established topoisomerase II poison m-AMSA is used for comparison. Results By utilizing a panel of Saccharomyces cerevisiae single-gene deletion strains, homologous recombination was identified as the most important DNA repair pathway determining the sensitivity towards ICRF-187. However, sensitivity towards m-AMSA depended much more on this pathway. In contrast, disrupting the post replication repair pathway only affected sensitivity towards m-AMSA. Homologous recombination (HR) defective irs1SF chinese hamster ovary (CHO) cells showed increased sensitivity towards ICRF-187, while their sensitivity towards m-AMSA was increased even more. Furthermore, complementation of the XRCC3 deficiency in irs1SF cells fully abrogated hypersensitivity towards both drugs. DNA-PK cs deficient V3-3 CHO cells having reduced levels of non-homologous end joining (NHEJ) showed slightly increased sensitivity to both drugs. While exposure of human small cell lung cancer (SCLC) OC-NYH cells to m-AMSA strongly induced γH2AX, exposure to ICRF-187 resulted in much less induction, showing that ICRF-187 generates fewer DNA double strand breaks than m-AMSA. Accordingly, when yeast cells were exposed to equitoxic concentrations of ICRF-187 and m-AMSA, the expression of DNA damage-inducible genes showed higher levels of induction after exposure to m-AMSA as compared to ICRF-187. Most importantly, ICRF-187 stimulated homologous recombination in SPD8 hamster lung fibroblast cells to lower levels than m-AMSA at all cytotoxicity levels tested, showing that the mechanism of action of bisdioxopiperazines differs from that of classical topoisomerase II poisons in mammalian cells. Conclusion Our results point to important differences in the mechanism of cytotoxicity induced by bisdioxopiperazines and topoisomerase II poisons, and suggest that bisdioxopiperazines kill cells by a combination of DNA break-related and DNA break-unrelated mechanisms.	Background Type II topoisomerases are essential nuclear enzymes found in all living organisms [ 1 ]. Their basic role in cells is to catalyse the transport of one DNA double helix through a transient double strand break in another DNA molecule [ 2 ]. This activity helps relieve tensions built up in DNA during various DNA metabolic processes such as DNA replication, chromosome condensation and de-condensation, chromosome segregation and transcription [ 3 ]. Topoisomerase II is also a major drug target in human cancer therapy, where a number of clinically active drugs such as the epipodophyllotoxins VP-16 and VM-26, the aminoacridine m-AMSA, and antracyclines such as doxorubicin, daunorubicin and epirubicin are widely used. These drugs have collectively been called topoisomerase II poisons due to their mechanism of action on topoisomerase II. Rather than inhibiting the basic catalytic activity of the enzyme, these drugs perturb the topoisomerase II catalytic cycle resulting in an increase in the level of a transient reaction intermediate, where DNA is cleaved and covalently attached to DNA [ 4 ]. Catalytic inhibitors of topoisomerase II have a different mode of action. These drugs exemplified by merbarone, aclarubicin, F11782 and the bisdioxopiperazines work by inhibiting topoisomerase II at other stages in the reaction cycle where DNA is not cleaved as reviewed in [ 5 , 6 ]. Amongst these, the bisdioxopiperazines have gained much attention due to their distinct and well-characterised mode of action. These compounds exemplified by ICRF-187, ICRF-159 and ICRF-154 inhibit the DNA strand passage reaction of topoisomerase II by sequestering this protein as a salt-stable closed clamp on DNA whose formation depends on the presence of ATP [ 7 - 9 ]. This closed clamp complex has retained the capability to hydrolyse ATP, although at a reduced level [ 10 ]. Several studies indicate that the closed clamp complex on DNA represents a novel form of DNA lesion to cells, – and that inhibition of topoisomerase II catalytic activity (DNA strand passage activity) is not responsible for bisdioxopiperazine-induced cell kill: ( i ) Expression of bisdioxopiperazine-sensitive topoisomerase II in cells also expressing bisdioxopiperazine-resistant topoisomerase II confers dominant sensitivity to these drugs [ 7 , 11 ] – a modality reminiscent of that of topoisomerase II poisons. ( ii ) Mouse embryonic stem cells [ 12 ] and chicken lymphoma DT40 cells [ 13 ] having one topoisomerase II α allele knocked out with concomitant reduced levels of topoisomerase II, are resistant to both ICRF-193 and the topoisomerase II poison etoposide, – while the opposite result is to be expected if ICRF-193 kill cells by depriving them of essential topoisomerase II catalytic activity. ( iii ) Killing of yeast cells by exposure to ICRF-193 occurs more rapidly and to a higher level than killing of yeast cells induced by the depletion of endogenous topoisomerase II catalytic activity [ 7 ]. ( iv ) The ICRF-193-induced topoisomerase II closed clamp complexes on DNA work as a "road block" signalling selective degradation of topoisomerase II β as well as p53 activation in a transcription dependent fashion [ 14 ]. Some studies have recorded elevated levels of DNA breaks in cells after exposure to the bisdioxopiperazine analog ICRF-193. In one study, ICRF-193 was found to increase the level of topoisomerase II-DNA covalent complexes in vitro and in vivo [ 15 ]. However, in this study efficient trapping of this covalent intermediate was only evident when guanidine was used to denature topoisomerase II attracted to DNA, while the agent normally used to trap the topoisomerase II-DNA cleavage complex, SDS, was not effective. In another study, the comet assay and pulsed field gel electrophoresis were used to demonstrate elevated levels of DNA breaks in mammalian cells after exposure to ICRF-193 [ 16 ]. In this study, inhibiting DNA replication with aphidicolin reduced the level of DNA breaks induced by the topoisomerase II poison m-AMSA, but had no effect on DNA breaks induced by ICRF-193. These results point towards bisdioxopiperazines poisoning DNA topoisomerase II in cells by a mechanism different from that of the classical topoisomerase II poisons such as etoposide and m-AMSA. In a recent paper, it was directly demonstrated that m-AMSA-induced dominant cytotoxicity only required the DNA cleavage activity of topoisomerase II, while dominant cytotoxicity towards ICRF-193 depended strictly on the DNA strand passage reaction of the enzyme[ 17 ]. Based on these observations, the present study aims to further elucidate the mechanism of cytotoxicity induced by the bisdioxopiperazines. We here characterise the effect of the clinically approved analog ICRF-187 (dexrazoxane) by using a number of different cell-based pharmacological assays, taking advantage of genetically modified yeast and mammalian cells. Results ICRF-187 sensitivity of yeast cells depends on their homologous recombination status, albeit to a lesser extent than for m-AMSA sensitivity To pinpoint the mechanism of cytotoxicity of ICRF-187 versus m-AMSA, we employed a panel of human topoisomerase II α-transformed haploid single-gene knockout yeast strains, defective in various aspects of DNA repair, checkpoint control, membrane transport and protein degradation. All yeast strains are depicted in table 1 . We used doses of these two drugs equitoxic to wild-type cells having no mutations. Clonogenic survival of all yeast strains is depicted in additional file 1 , and the degree of drug resistance / hypersensitivity is also listed in table 2 . Table 1 Yeast strains used in the study BY4741 + pMJ1 BY4741Δ rad9 + pMJ1 BY4741Δ rad51 + pMJ1 BY4741Δ tel1 + pMJ1 BY4741Δ rad52 + pMJ1 BY4741Δ chk1 + pMJ1 BY4741Δ rad54 + pMJ1 BY4741Δ mhl1 + pMJ1 BY4741Δ rad55 + pMJ1 BY4741Δ pms1 + pMJ1 BY4741Δ rad57 + pMJ1 BY4741Δ msh2 + pMJ1 BY4741Δ rad59 + pMJ1 BY4741Δ msh3 + pMJ1 BY4741Δ dcm1 + pMJ1 BY4741Δ atr1 + pMJ1 BY4741Δ sae2 + pMJ1 BY4741Δ pdr5 + pMJ1 BY4741Δ rad50 + pMJ1 BY4741Δ yor1 + pMJ1 BY4741Δ mre11 + pMJ1 BY4741Δ ubc4 + pMJ1 BY4741Δ xrs2 + pMJ1 BY4741Δ ubc13 + pMJ1 BY4741Δ rad6 + pMJ1 BY4741Δ doa4 + pMJ1 BY4741Δ rad18 + pMJ1 BY4741Δ qri8 + pMJ1 BY4741Δ rev1 + pMJ1 BY4741Δ rnr3 + pMJ1 BY4741Δ rev3 + pMJ1 BY4741Δ sml1 + pMJ1 BY4741Δ rad1 + pMJ1 BY4741 + PYX112 BY4741Δ rad14 + pMJ1 BY4741Δ rad6 + pYX112 BY4741Δ apn1 + pMJ1 BY4741Δ rad50 + pYX112 BY4741Δ yku70 + pMJ1 BY4741Δ rad52 + pYX112 BY4741Δ yku80 + pMJ1 BY4741Δ sae2 + pYX112 BY4741Δ mec3 + pMJ1 BY4741Δ yku70 + pYX112 BY4741Δ dcc1 + pMJ1 BY4741Δ rad17 + pMJ1 JN362A t2-4 + pMJ1 Table 2 Hypersensitivity (or resistance) scoring of pMJ1-transformed BY4741 deletion strains towards ICRF-187 and m-AMSA as determined in clonogenic assay using 22.5 hours drug exposure. Drug ICRF-187 m-AMSA Gene deleted WT 0 0 Nucleotide Excision Repair (NER) Single Strand Annealing (SSA) Recombination Anti-Recombination Δ rad1 R 0 Δ rad14 0 0 Mismatch Repair (MMR) Anti-Recombination Δ msh2 R 0 Δ msh3 0 0 Δ mhl1 R 0 Δ pms1 R 0 Base Excision Repair (BER) Δ apn1 0 0 Post Replication Repair (PRR) Δ rev1 0 0 Δ rev3 0 0 Δ rad18 0 + Δ rad6 0 ++ Homologous Recombination (HR) Non-Homologous End Joining (NHEJ) Δ rad50 + ++ Δ mre11 + ++ Δ xrs2 + ++ Homologous Recombination (HR) Δ rad52 + ++ Δ rad51 + + Δ rad54 + ++ Δ rad57 + ++ Δ rad55 + ++ Δ rad59 0 0 Δ dmc1 0 0 Δ sae2 + + Non-Homologous End Joining (NHEJ) Δ yku70 + 0 Δ yku80 0 0 DNA Damage Checkpoints Δ tel1 0 0 Δ rad9 0 0 Δ mec3 0 0 Δ ddc1 0 0 Δ rad17 0 0 Δ chk1 0 0 ABC Transporters (Yeast MDR1 homologous) Δ atr1 0 0 Δ pdr5 + 0 Δ yor1 0 0 Ubiquitin conjugation / hydrolysis Δ ubc4 0 0 Δ ubc13 0 0 Δ doa4 0 R Δ qri8 0 0 Ribonucleotide-reductase regulation Δ rnr3 0 0 Δ sml1 0 0 R : Cells are more than a 1/2 log resistant at any drug concentration. 0 : Cells are no more than a 1/2 log resistant and no more than a 1/2 log hypersensitive at any drug concentration. + : Cells are at least a 1/2 log but no more than 2 log hypersensitive at any concentration. ++ : Cells are more than 2 log hypersensitive at any concentration. Hypersensitivity (resistance) was graduated as follows The products of the three genes RAD50 , MRE11 and XRS2 together form the Rad50/Mre11/Xrs2 hetero-trimer protein complex that has catalytic and structural functions in many kinds of DNA metabolic processes including HR as reviewed in [ 18 ]. We observed that Δ rad50 , Δ mre11 , and Δ xrs2 single knockout strains were extremely hypersensitive towards m-AMSA, while they displayed considerably less hypersensitivity towards ICRF-187 ( additional file 1 and table 2 ). We also tested the effect of deleting a number of genes exclusively involved in HR namely RAD51 , RAD52 , RAD54 , RAD55 , RAD57 , RAD59 , DMC1 and SAE2 [ 18 ]. Deleting RAD51 , RAD52 , RAD54 , RAD55 , RAD57 and SAE2 had a profound effect on the sensitivity of the yeast cells towards m-AMSA while having a smaller, but significant, effect on the sensitivity of these cells towards ICRF-187 ( additional file 1 and table 2 ), again pointing to the HR pathway as being most important for the repair of DNA damage caused by cleavage complex stabilising drugs. We found that deleting RAD59 had no effect on drug sensitivity, confirming reported data that RAD59 only becomes functionally important in the absence of functional Rad51 protein [ 19 ]. We also observed no effect of deleting DMC1 ( additional file 1 and table 2 ). This may be explained by the fact that Dmc1p is primarily involved in meiotic recombination [ 20 ]. NHEJ represents another DNA repair-pathway. In yeast, this repair pathway is generally less important than HR for the repair of DNA breaks [ 21 ]. In accordance with this we observed no effect of deleting the NHEJ genes YKU70 and YKU80 on the sensitivity towards m-AMSA. We did however, observe some hypersensitivity of Δ yku70 cells towards ICRF-187, while Δ yku80 cells were not hypersensitive ( additional file 1 and table 2 ). This is a surprising result, because Yku70p and Yku80p have been demonstrated to play equally important roles for NHEJ activity in yeast [ 21 ]. These results suggest that the effect of deleting YKU70 is unrelated to its DNA repair functions. The DNA binding Rad18p forms a hetero-dimer with Rad6p that is involved in post replication repair (PRR) [ 22 ]. We found that although Δ rad18 cells were clearly hypersensitive towards m-AMSA, Δ rad6 cells were markedly more sensitive. Δ rad6 cells were actually among the most sensitive towards m-AMSA (figure 1 , additional file 1 and table 2 ). Interestingly, the sensitivity of Δ rad6 and Δ rad18 cells towards ICRF-187 is indistinguishable from that of wild-type cells (figure 1 ). The vast difference in the sensitivity of Δ rad 6 cells towards ICRF-187 and m-AMSA confirms the notion that the DNA lesions induced by these drugs are different in nature. The finding that Δ rad6 cells are much more sensitive towards m-AMSA than Δ rad18 cells is surprising, and may indicate that Rad6p functions unrelated to DNA repair affect cellular sensitivity towards m-AMSA. Rad6p has ubiquitin conjugating activity [ 22 ], and therefore such Rad18p-unrelated functions could involve protein degradation via the 26S proteasome pathway. To test this hypothesis, we analysed the drug sensitivity of four yeast strains with impaired protein degradation; Δ ubc4 , Δ ubc13 , Δ doa4 and Δ qri8 . These deletion strains were not hypersensitive towards m-AMSA (or ICRF-187) ( additional file 1 and table 2 ), suggesting that Δ rad6 cells are hypersensitive towards m-AMSA due to impaired PRR activity. The involvement of both HR repair and PRR in determining the sensitivity of yeast cells towards the topoisomerase II cleavage complex stabilising drugs mitoxantrone and idarubicin has previously been reported [ 23 ]. The observed lack of hypersensitivity of the Δ rev1 and Δ rev3 strains ( additional file 1 and table 2 ) suggests that trans-lesion DNA synthesis plays no role in determining the sensitivity towards ICRF-187 or m-AMSA. Figure 1 Clonogenic sensitivity of PRR defective Δ rad6 and Δ rad18 yeast cells towards equitoxic doses of ICRF-187 and m-AMSA. A Δ rad52 strain is included for comparison. Error-bars represent SEM of at least 3 experiments. We also analysed the effect of deleting genes belonging to the nucleotide excision repair (NER) pathway – RAD1 and RAD14 , the base excision repair (BER) pathway – APN1 , and the mismatch repair (MMR) pathway – MLH1 , PMS1 , MSH2 and MSH3 . None of these deletions caused cells to become more sensitive towards ICRF-187 and m-AMSA, indicating that these pathways are not involved in repairing DNA damage induced by these drugs. Interestingly, deleting genes involved in the MMR and NER pathways caused cells to become somewhat resistant to ICRF-187, and to a lesser extent towards m-AMSA ( additional file 1 and table 2 ). The products of these genes have been implicated to have anti-recombination activities [ 24 ]. Increased levels of recombination in these cells could therefore be responsible for the observed low-level resistance towards ICRF-187. Deletion of DNA damage checkpoint genes has little effect on both ICRF-187 and m-AMSA sensitivity of yeast cells While deleting genes involved in DNA repair caused cells to be hypersensitive towards both drugs tested, we observed little effect of deleting the DNA damage checkpoint genes MEC3 , DDC1 , RAD17 , TEL1 , RAD9 and CHK1 ( additional file 1 and table 2 ). Our finding that checkpoint control regulation plays no important role for bisdioxopiperazine sensitivity supports earlier data showing that arresting yeast cells in G1 phase did not protect against ICRF-193 cytotoxicity [ 7 ]. The lack of importance of checkpoint function in determining sensitivity towards m-AMSA is also in accordance with published observations [ 23 ], where sensitivity towards the cleavage complex stabilising topoisomerase II poisons mitoxantrone, idarubicin, daunorubicin and doxorubicin were only marginally affected by inactivating the RAD9 , RAD17 , MEC1 , and RAD53 genes, while the sensitivity of yeast cells towards the topoisomerase I poison camptothecin showed a strong dependency on these pathways. ICRF-187 is a possible substrate for the Pdr5 ABC transporter in yeast In mammalian cells resistance towards various structurally-unrelated anti-neoplastic agents is often associated with over-expression of ABC-type drug efflux transporters such as p-glycoprotein and multi-drug resistance protein (MRP) as reviewed in [ 25 ]. Among the three yeast ABC transporters assessed in our study, Pdr5 is by far the best characterised [ 26 ]. While deleting YOR1 and ATR1 had no effect on drug sensitivity, Δ pdr5 cells were clearly hypersensitive towards ICRF-187 but not towards m-AMSA ( additional file 1 and table 2 ), suggesting that ICRF-187 is a substrate for the Pdr5 pump in yeast. It has to be emphasized, that over-expression of drug efflux pumps has not been associated with resistance towards bisdioxopiperazines in mammalian cells. Transcriptional profiling of yeast cells after exposure to equitoxic concentrations of ICRF-187 and m-AMSA In order to assess the effect on global gene expression of the interaction between human topoisomerase II α and the two drugs in yeast cells, transcriptional profiling was performed using Affymetrix gene chip technology. We exposed pMJ1-transformed JN362A t2–4 yeast cells expressing human topoisomerase II α as their sole active topoisomerase II to equitoxic doses of ICRF-187 and m-AMSA for two hours at 34°C. This treatment resulted in a 50 % reduction in clonogenic survival after exposure to both drugs ( additional file 2 ). Genes whose average expression in two independent experiments was up- or down-regulated more than 1.5 fold by exposure to the drugs were filtered out. 138 transcripts were induced by exposure to ICRF-187 while the number was 90 for m-AMSA. 26 transcripts were repressed by exposure to ICRF-187 while the number was 16 for m-AMSA. Additional file 3 lists transcripts induced or repressed by ICRF-187 while additional file 4 lists transcripts induced or repressed by m-AMSA. The expression profile of selected genes is listed in Table 3 and discussed below. Table 3 Selected genes induced or repressed by exposure of pMJ1-transformed yeast cells to equitoxic concentrations of ICRF-187 and m-AMSA for 2 hours Transcriptional activation Gene function Gene Name Effect of ICRF-187 Effect of m-AMSA DNA damage RNR3 3.2 4.0 HUG1 3.1 5.0 RAD51 1.9 2.0 RAD54 1.7 1.6 RNR2 1.5 1.8 Membrane transport PDR12 2.4 1.0 PDR15 2.0 1.0 Stress response HSP12 2.8 1.6 HSP26 1.7 1.8 WSC4 1.6 1.3 XBP1 1.6 1.5 HSP42 1.5 1.5 Others SWI1 1.6 1.1 Transcriptional repression Others SNZ1 0.6 1.1 PCL9 0.7 0.6 Genes induced by both drugs Both compounds induced the expression of a number of genes known to be up-regulated by DNA damage. The expression of four well-established DNA-damage inducible genes; RNR2 , RNR3 [ 27 , 28 ] and RAD51 , RAD54 [ 29 ] was thus induced by both drugs. Both compounds also stimulated the expression of HUG1 recently shown to be up-regulated by DNA damage and replication arrest [ 30 ] (See Table 3 , additional files 3 and 4 ). Furthermore, both drugs stimulated expression of the stress-inducible XBP1 gene whose protein product is a transcription factor. XBP1 expression is reportedly induced in response to heat shock, high osmolarity, oxidative stress, glucose starvation and DNA damage, and induces a slow-growth phenotype with lengthening of the G1 cell cycle phase [ 31 ]. The PCL9 gene product has cyclin-dependent protein kinase regulator activity suggesting a role for Pcl9p in cell cycle regulation [ 32 ]. Repression of PCL9 by exposure to both drugs (table 3 , additional files 3 and 4 ) may thus be indicative of drug-induced cell cycle arrest in accordance with the XBP1 expression data. Finally, both drugs induced the expression of general stress-induced HSP genes as expected (table 3 , additional files 3 and 4 ). Although expression of the RNR3 and HUG1 genes was up-regulated by both drugs, pMJ1-transformed cells having RNR3 or SML1 deleted (the latter is a functional non-inducible homolog of HUG1 ) have wild-type sensitivity towards both drugs (table 2 , additional file 1 ), showing that although these genes are induced by both drugs, they are probably not involved in determining their cytotoxicity. Genes specifically induced by ICRF-187 We found that ICRF-187 specifically induced the expression of two genes encoding the ABC efflux transporters Pdr12 and Pdr15, while m-AMSA had no effect on the expression of these genes (table 3 , additional files 3 and 4 ). Transcription of the stress-inducible WSC4 gene was likewise enhanced by exposure to ICRF-187 (table 3 , additional files 3 and 4 ). Knocking out WSC4 in yeast cells has been found to enhance their sensitivity towards various stresses including heat, ethanol and DNA damage [ 33 ]. Recently, the SWI/SNF complex was directly shown to repress transcription in S. cerevisiae cells [ 34 ]. We found that SWI1 was specifically induced by ICRF-187 (table 3 , additional files 3 and 4 ). Finally, we found that ICRF-187 specifically repressed the expression of the stationary phase-induced SNZ1 gene [ 35 ] (table 3 , additional files 3 and 4 ). Exposure of yeast cells to ICRF-187 causes less transcriptional induction of DNA damage-inducible genes than exposure to m-AMSA at equitoxic drug concentrations To verify the array data we performed real-time PCR to assess the expression of the RNR3 , HUG1 , RAD51 and RAD54 genes after exposure to the two drugs using the actin gene ACT1 as internal control (figure 2 ). Real-time PCR confirmed induction of these established DNA damage-inducible genes by both drugs assessed. Furthermore, exposure of the cells to m-AMSA resulted in a higher level of induction than did exposure to ICRF-187 for the four genes tested, especially for HUG1 . These data suggest that when yeast cells are exposed to equitoxic concentrations of the two drugs, m-AMSA generates more extensive DNA damage than ICRF-187. Figure 2 Analysis of gene expression by real time PCR. Real-time PCR was used to determine the expression of the DNA-damage inducible genes RNR3 , HUG1 , RAD51 , and RAD54 by using the 2- ΔΔCt method. Gene expression was normalized to that of the actin gene ACT1 . It can be seen that exposure of yeast cells to m-AMSA results in higher levels of induction of transcription of these four genes than exposure to ICRF-187 when the two drugs are applied at equitoxic concentrations. Error-bars represent SEM of two independent experiments each performed in duplicate. ICRF-187 sensitivity of mammalian hamster cells depends on their homologous recombination status, albeit to a lesser extent than seen for m-AMSA sensitivity The yeast clonogenic assays presented above point to an important role of HR in the repair of m-AMSA-induced DNA damage, while the importance of this pathway in the repair of ICRF-187-induced DNA damage is less so. Because HR is the major repair pathway in yeast [ 21 ], while both NHEJ and HR are important for the repair of DNA breaks in mammalian cells [ 36 ], we next turned to assess the importance of these pathways in mammalian cells having reduced levels of HR and NHEJ. In this analysis we used a panel of four hamster cell lines; AA8 cells (wild-type), irs1SF cells [ 37 ] (recombination defective caused by non-functional XRCC3), CXR3 cells [ 37 ] (recombination proficient due to ectopic expression of human XRCC3), and V3-3 cells [ 38 ] (reduced level of NHEJ due to non-functional DNA-PK cs ). We observed a strong dependence on HR for the sensitivity towards m-AMSA (figure 3A ). Thus, only 1 % relative survival was seen for the irs1SF cells (recombination defective) at 6 nM of this drug, while wild-type AA8 cells were only slightly sensitive to15 nM m-AMSA. Furthermore, ectopic expression of human XRCC3 fully reversed the m-AMSA hypersensitivity as CXR3 cells were no more hypersensitive than AA8 wild-type cells, confirming the notion that HR plays a role in the repair of topoisomerase II-induced DNA breaks in mammalian cells. We also observed that irs1SF cells were hypersensitive towards ICRF-187 (figure 3B ), but the degree of hypersensitivity was much less than observed for m-AMSA, as also seen for recombination deficient yeast cells. Again, ectopic expression of the human XRCC3 homolog reversed the hypersensitivity as CXR3 cells displayed near wild-type sensitivity towards ICRF-187. Figure 3 Assessing the clonogenic sensitivity of HR and NHEJ deficient and proficient hamster cells towards ICRF-187 and m-AMSA. To determine the sensitivity of the four cell lines AA8 (wild-type), irs1SF (recombination defective caused by non-functional XRCC3), CXR3 (recombination proficient due to ectopic expression of human XRCC3), and V3-3 (defective in NHEJ due to non-functional DNA-PK cs ) towards ICRF-187 and m-AMSA, a clonogenic assay with continuous drug exposure was used. Error-bars represent SEM of two independent experiments. DNA-PK cs deficient hamster cells show slightly increased sensitivity towards both m-AMSA and ICRF-187 To assess the effect of NHEJ on drug sensitivity we also employed the V3-3 cell line (DNA-PK cs deficient, with concomitant reduced level of NHEJ). The results of these experiments are depicted in figure 3A and 3B . The V3-3 cells were slightly hypersensitive towards both drugs suggesting a role for NHEJ in the repair of DNA lesions induced by both drugs. AA8, irs1SF, CXR3 and V3-3 cells have similar levels of topoisomerase II catalytic activity The sensitivity of cells towards topoisomerase II directed drugs depends both on their levels of topoisomerase II catalytic activity, and on their capability to repair topoisomerase II-induced DNA damage. We therefore determined the level of topoisomerase II catalytic (DNA strand passage) activity in crude protein extracts isolated from the four cell lines used in clonogenic assays, by applying a radioactive decatenation assay. No significant difference in the level of topoisomerase II DNA strand passage activity was recorded between the four cell lines ( additional file 5 ). This result rules out the possibility that varying levels of topoisomerase II catalytic activity in these cells is responsible for their differential drug sensitivity. ICRF-187 induces lower levels of homologous recombination in hamster cells than m-AMSA at equitoxic concentrations The hypersensitivity of the recombination defective irs1SF cells towards both drugs suggests that HR is involved in repairing DNA lesions induced by both drugs. To address this directly we applied a mammalian recombination assay to measure stimulation of HR by ICRF-187 and m-AMSA by using SPD8 hamster cells [ 39 ]. This assay measures the repair of a defective chromosomal hprt gene by the activity of HR. From figure 4A it is evident that both drugs stimulated the level of HR in a dose dependent manner. When recombination frequency is expressed as a function of surviving cells (figure 4C ) it becomes evident that the recombination frequency increases with increasing cell mortality for both drugs tested. From figure 4C it is also evident that at equitoxic concentrations of the two drugs, m-AMSA stimulated HR to much higher levels than did ICRF-187. Thus, at 50 % survival, no induction of HR was seen with ICRF-187 (in three independent experiments), while m-AMSA caused an approximately 10-fold induction at equitoxic doses. Figure 4 Assessing the effect of equitoxic concentrations of ICRF-187 and m-AMSA on the level of HR in SPD8 hamster cells. The SPD8 cell line has a defective hprt gene that can be repaired by HR. Panel A depicts the induction of HR induced by increasing concentrations of the two drugs. Panel B depicts the relative survival of cells exposed to similar concentrations of the two drugs. The data represented in Panel A and B was used to generate Panel C, where recombination frequency is plotted against the surviving fraction of cells. This data presentation allows a direct comparison of recombination levels at equitoxic concentrations of the two drugs. Representative data from one of three independent experiments is shown. ICRF-187 induces only low levels of H2AX phosphorylation in human SCLC cells as compared to m-AMSA Induction of γH2AX is a well-established marker for topoisomerase-induced DNA double strand breaks in mammalian cells [ 40 - 42 ]. We therefore assessed the effect of exposing human SCLC OC-NYH cells to 10 μM m-AMSA and 1 mM of ICRF-187 at increasing time points (figure 5A and 5B ). Exposure to 10 μM m-AMSA quickly resulted in γH2AX induction. Thus, induction was evident after 30 min, and after 24 hours more than 10-fold induction was observed. In contrast, when cells were exposed to 1 mM ICRF-187, much less γH2AX induction was observed, and after 24 hours the level of induction was less that three-fold. Figure 5 Assessing the effect of m-AMSA and ICRF-187 on γH2AX induction in human SCLC OC-NYH cells. To assess the level of DNA breaks in human cells after exposure to 10 μM m-AMSA and 1 mM ICRF-187 for increasing time points, total histones were isolated after incubation with the drugs. 10 μg of purified histones was then used in western blotting experiments. Panel A depicts a typical Western blot showing increased γH2AX induction with increasing drug incubation times. Panel B depicts fold γH2AX induction plotted against drug incubation time to analyse the kinetics of induction of DNA double strand breaks by the two drugs. Insert shows γH2AX induction from 0 to 2 hours for better resolution. Error-bars represent SEM of three independent experiments for ICRF-187 treatment and SEM of two independent experiments for m-AMSA treatment. Discussion We initiated the study by assessing the clonogenic sensitivity of yeast single-gene deletion mutants ectopically expressing human topoisomerase II α towards m-AMSA and ICRF-187. The results presented in table 2 and additional file 1 indicates that HR plays a role in the repair of ICRF-187-induced DNA damage. Previous studies addressing the bisdioxopiperazine sensitivity of yeast cells have generated different results. In one study rad52 -cells had the same sensitivity towards ICRF-187 and ICRF-193 as did RAD52 + cells [ 7 ], while in other studies, HR deficient cells were found to be hypersensitive towards bisdioxopiperazines, although to a much lesser extent than towards topoisomerase II cleavage complex stabilising drugs [ 43 , 44 ]. Our present study involving numerous other genes involved in various aspects of HR clearly establishes this pathway as being a functional determinant for bisdioxopiperazine sensitivity in yeast cells. In a recent work by Simon and colleagues, where a panel of yeast deletion strains was also applied to pinpoint the mechanism of action of various anticancer drugs, a given drug was classified as selective if one single pathway was mainly involved in determining cellular sensitivity [ 23 ]. The selective involvement of the HR pathway in determining the sensitivity towards ICRF-187 classifies this drug as highly selective according to this definition. However, it is important to note that although HR clearly does play a role in protecting yeast cells from ICRF-187 cytotoxicity, the importance of this pathway on cell survival in the presence of m-AMSA is much greater ( additional file 1 , table 2 ) – in accordance with this drug being a topoisomerase II poison killing cells solely by the generation of topoisomerase II-mediated DNA breaks. We find that the relative sensitivity of AA8, irs1SF, and CXR3 cells towards m-AMSA (figure 3 ) closely resembles their sensitivity towards etoposide [ 45 ], showing that the hypersensitivity of XRCC3 deficient irs1SF cells is general to topoisomerase II poisons, suggesting a role for HR in the repair of topoisomerase II-induced DNA breaks in these cells. The involvement of HR in the repair of DNA lesions induced by topoisomerase II poisons in higher eukaryotes is also supported by a recent work suggesting that RAD51 plays an important role in the repair of etoposide-induced DNA damage in human small cell lung cancer cells [ 46 ], and by work by Adachi and colleagues who recently found that knocking out RAD54 in chicken DT40 cells enhances their sensitivity towards the topoisomerase II poison etoposide [ 13 ]. We find that XRCC3 defective irs1SF cells are more sensitive towards ICRF-187 than the parental AA8 cells, although the XRCC3 defect has a much more pronounced effect on m-AMSA sensitivity (figure 3A versus figure 3B ) – as seen with the yeast deletion mutant panel. Adachi and colleagues have found that knocking out RAD54 in DT40 chicken cells does not increase sensitivity towards ICRF-193 [ 13 ]. The reason for this discrepancy in not clear. The difference observed between the DT40 and irs1SFcells may relate to the fact that different DNA repair genes are deleted in the two cell lines possibly resulting in different processing of bisdioxopiperazine-induced DNA damage. In any case, this discrepancy does not challenge the overall finding that HR plays a more important role in protecting cells of various origin from cytotoxicity induced by topoisomerase II poisons as compared to cytotoxicity induced by bisdioxopiperazines. To study the importance of NHEJ in determining the sensitivity towards m-AMSA and ICRF-187 we employed a DNA-PK cs defective hamster cell line, V3-3, which has reduced levels of NHEJ activity. We find that V3-3 cells are hypersensitive towards m-AMSA (figure 3A ). This result is in accordance with a recent publication by Willmore and colleagues who found that a specific small-molecule inhibitor of DNA-PK cs NU7026 could potentate the sensitivity of human leukemic K562 cells towards various topoisomerase II poisons [ 47 ]. Our result is also in accordance with a recent report by Adachi and colleagues showing that DNA-PK cs knockout chicken DT40 cells are hypersensitive towards etoposide [ 48 ]. These result points towards an important role of DNA-PK cs in determining the sensitivity of higher vertebrate cells towards topoisomerase II poisons. Different studies have demonstrated a more pronounced effect of inactivating Ku as compared to DNA-PK cs on cellular sensitivity towards topoisomerase II poisons [ 48 - 50 ]. Consequently, the importance of NHEJ in determining the sensitivity towards topoisomerase II poisons in mammalian cells is likely to be underestimated from our V3-3 cell data. This notion is confirmed by early publications demonstrating that Ku deficient hamster cells are highly sensitive towards m-AMSA and etoposide [ 51 , 52 ]. Our finding that V3-3 cells are more sensitive towards ICRF-187 than AA8 cells (figure 3B ) is also in accordance with observations by Adachi and colleagues who find that DNA-PK cs deficient chicken DT40 cells are hypersensitive towards another bisdioxopiperazine analog, ICRF-193. These authors found the effect of inactivating DNA-PK cs to be much more pronounced than seen in our present study. While the reason for this difference is not clear, it has to be mentioned that studies addressing the effect of DNA-PK cs on the sensitivity towards topoisomerase II targeting drugs and ionising radiation have produced varying results. Thus, in a study by Jin and colleagues, DNA-PK cs defective murine cells were much less sensitive towards etoposide than Ku70 and Ku80 deficient cells [ 50 ], while in a study by Gao and colleagues the importance of DNA-PK cs on the sensitivity towards ionising radiation was found to depend on cell type and/or cell cycle distribution [ 49 ]. Such variation could well explain the different importance of DNA-PK cs observed in our study and in the work by Adachi and colleagues. In order to study directly the effect of exposing mammalian cells to ICRF-187 and m-AMSA on the levels of HR, we employed a mammalian recombination assay previously described [ 39 ]. In this assay, m-AMSA enhanced the level of recombination in SPD8 cells to higher levels than ICRF-187 at all cytotoxicity levels tested (figure 4C ), demonstrating directly pronounced differences in the mechanism(s) by which topoisomerase II poisons and bisdioxopiperazines kill cells. This notion is further confirmed by our γH2AX induction experiments, where ICRF-187 causes much lower levels of induction (figure 5 ), demonstrating that ICRF-187 induces less DNA breaks in cells than m-AMSA. Our observation that ICRF-187 induces both HR and γH2AX induction in mammalian cells, is in agreement with a recent paper demonstrating by the use of comet assay and pulsed field gel electroforesis that ICRF-193 induces DNA breaks in mammalian cells [ 16 ]. This result is also in agreement with our real-time PCR results where ICRF-187 tended to induce the expression of established DNA damage-inducible genes. The finding that ICRF-187 induces lower levels of HR than m-AMSA in SPD8 cells at equitoxic doses may be explained in at least two ways. Bisdioxopiperazine-induced DNA breaks could be more toxic to cells than breaks induced by topoisomerase II poisons, or the DNA breaks could be only partly responsible for killing the cells. Three lines of evidence support the latter possibility. ( i ) Functional ATR, but not ATM, is required for a cell cycle checkpoint arrest induced by ICRF-193 [ 53 ], suggesting that DNA breaks are not involved in triggering the checkpoint signal. ( ii ) Exposure of mammalian cells to the topoisomerase II poison etoposide induces degradation of the large subunit of RNA polymerase II indicative of DNA breaks, while this is not the case for ICRF-193 [ 14 ]. ( iii ) Our finding that cell survival in the presence of ICRF-187 depends less on HR than cell survival in the presence of m-AMSA suggests that ICRF-187-induced DNA breaks contribute less to overall cytotoxicity than m-AMSA-induced DNA breaks. If ICRF-187-induced DNA breaks were more toxic to cells than m-AMSA-induced DNA breaks, cell survival in the presence of ICRF-187 would be expected to depend at least as much on HR as cell survival in the presence of m-AMSA. This is not the case. What mechanisms are then responsible for producing the DNA breaks induced by bisdioxopiperazines in cells? In a recent work by Oestergaard and colleagues).)[ 17 ], it is suggested that the toxic intermediate causing bisdioxopiperazine cytotoxicity is topoisomerase II stably bound to two DNA segments – a conformation they suggested would only be attainable if the DNA strand passage reaction of topoisomerase II is functioning. HR could then be required for the repair of DNA breaks generated by the collision of DNA tracking complexes with such four-way DNA junctions / topoisomerase II closed clamp complexes on DNA. It has recently been demonstrated by the use of pulsed field gel electrophoresis, that inhibiting DNA replication by aphidicolin does not reduce the level of DNA breaks generated by exposure of mammalian cells to ICRF193, while the level of m-AMSA-induced DNA breaks was reduced by aphidicolin treatment [ 16 ]. This result suggests that collision of the DNA replication complex with bisdioxopiperazine-induced topoisomerase II closed clamp complex on DNA is not involved in generating the DNA breaks. In a recent study by Lundin and colleagues, it was demonstrated that inhibiting DNA replication by exposing cells to hydroxyurea resulted in the generation of DNA breaks [ 54 ]. Furthermore, in this work as well as in a subsequent work [ 45 ], HR was shown to be functionally involved in repairing such DNA breaks. The first of these two studies used the same four hamster cell lines that are also used in our present study. Remarkably, the relative sensitivity of these cell lines towards hydroxyurea exactly resembles their sensitivity towards ICRF-187 seen in our present work. This may suggest that replication arrest is involved in generating DNA breaks induced by bisdioxopiperazines in cells. Here, replication forks stalled at the bisdioxopiperazine-induced closed clamp complexes could be the source of DNA breaks in newly replicated DNA [ 54 ]. This would also explain the lack of effect of aphidicolin on the level of ICRF-193-induced DNA breaks observed by Hajji and colleagues [ 16 ]. If the DNA breaks result from arrested replications forks, and not from the collision of the DNA replication complex with the closed clamp complex on DNA, no effect of aphidicolin would be expected. This mechanism would also explain why yeast cells arrested in intra-S phase are not protected from ICRF-193 cytotoxicity [ 7 ]. We therefore suggest that this mechanism is responsible for generating DNA breaks induced by bisdioxopiperazines in cells. Together our HR, γH2AX, and cytotoxicity data suggest that bisdioxopiperazines kill cells by a combination of DNA break-related and DNA break-unrelated mechanisms. This raises the question as to which mechanism(s) is / are involved in mediating the DNA break-unrelated part of bisdioxopiperazine cytotoxicity. Exposure of mammalian cells to ICRF-193 represses global transcription and mediates selective degradation of topoisomerase II β via a transcription dependent mechanism [ 14 ]. Inhibition of the RNA polymerase II – transcription complex by bisdioxopiperazine-induced topoisomerase II complexes on DNA could therefore be involved in mediating the DNA break-unrelated component of bisdioxopiperazine cytotoxicity. Treatment of mammalian cells with high doses of ICRF-187 for one hour is capable of antagonising DNA breaks and the cytotoxicity of topoisomerase II poisons [ 55 , 56 ], and this antagonism can be extended to animal models, where ICRF-187 can antagonise etoposide toxicity [ 57 , 58 ] and bone marrow depression (unpublished results). How are bisdioxopiperazines capable of antagonising the effects of topoisomerase II while at the same time producing DNA breaks? Two independent studies assessing the dose- and schedule-dependency of combinations of bisdioxopiperazines and topoisomerase II poisons on cytotoxicity in mammalian cells may provide important clues. One study investigated the effect of combinations of ICRF-193 and etoposide [ 59 ]. Here, continuous administration of low doses of both drugs resulted in synergistic cell kill, while treatment with high concentrations of ICRF-193 for one hour efficiently antagonised etoposide-mediated cytotoxicity. A similar effect of schedule and concentration on cytotoxicity has also been observed for combinations of ICRF-187 and daunorubicin [ 60 ], but here long time exposure of the cells to both drugs resulted in an additive effect on cell kill. We have previously shown that exposure of mammalian cells to high concentrations of ICRF-187 (500 – 1000 μM) alone for 60 min is non-toxic, and that this treatment efficiently antagonises etoposide-induced DNA breaks and cytotoxicity [ 61 , 62 ]. In these studies, exposure of cells to 200 μM ICRF-187 was found to trap most cellular topoisomerase II α and β as non-extractable complexes on DNA. The inability of topoisomerase II poisons to act on bisdioxopiperazine-stabilised closed clamp complexes on DNA could therefore explain the antagonistic effect of high concentrations of bisdioxopiperazines generally observed in one-hour drug exposure experiments [ 59 - 62 ]. When a low concentration of bisdioxopiperazine is administered, it is most likely that only a small fraction of the topoisomerase II molecules in the cell is trapped as closed clamp complexes on DNA, leaving some or most topoisomerase II molecules available for the action of topoisomerase II poisons. Therefore, after long-time exposure of cells to low concentrations of bisdioxopiperazine and a topoisomerase II poison, covalent and non-covalent complexes of topoisomerase II on DNA could both contribute to cytotoxicity by generating DNA breaks via different mechanisms, thus explaining the additive or synergistic effect on cell kill observed under these circumstances. To summarise, our data are consistent with a model where bisdioxopiperazine-induced cytotoxicity results from a combination of DNA break-related and -unrelated mechanisms, where the DNA-break unrelated mechanism is clearly not mediated by the inhibition of catalytic topoisomerase II activity in the cells. Conclusion Since the discovery by Andoh and colleagues in 1991, that the bisdioxopiperazines target eukaryotic topoisomerase II [ 63 , 64 ], their mode of cytotoxicity has been the cause of debate. While early publications tended to classify these compounds as "pure" catalytic inhibitors of topoisomerase II, expected to kill cells by depriving them of essential topoisomerase II catalytic activity, numerous recent reports present data that are not consistent with this view [ 7 , 11 - 14 , 16 , 17 ]. In the present report we have characterised bisdioxopiperazine (ICRF-187) induced cytotoxicity in yeast and mammalian cells by using a combination of genetic and molecular approaches. Our results are consistent with a model where bisdioxopiperazines cause cytotoxicity by stabilising a topoisomerase II reaction intermediate / complex on DNA inducing DNA breaks in cells which are repaired by HR and NHEJ. We propose that cells exposed to bisdioxopiperazines die by a combination of DNA break-related and-DNA break-unrelated mechanisms. Our study clearly establishes that bisdioxopiperazines do not kill cells solely by depriving them of topoisomerase II catalytic activity. Methods Drugs ICRF-187 (Cardioxane, Chiron group) was dissolved in sterile water at 20 mg/ml and kept at – 80°C. To avoid hydrolysis of the drug, fresh aliquots were used for each experiment. m-AMSA (Pfizer) was diluted in DMSO and stored at – 80°C at 1 mg/ml. L-azaserine and thymidine (both from Sigma) were added directly to tissue culture medium. 6-thioguanine and hypoxanthine (both from Sigma) were dissolved in 5 M NaOH and immediately added to the tissue culture medium. Yeast strains and constructs BY4741 haploid Saccharomyces cerevisiae cells ( MAT a his3 Δ 1 leu2 Δ 0 met15 Δ 0 ura3 Δ 0 ) and a panel of single-gene deletion derivatives hereof (table 1 ) were purchased from EUROSCARF, Institute of Microbiology, Johann Wolfgang Goethe University Frankfurt, Germany. The construction of BY4741 and its deletion derivatives have been described [ 65 ]. JN362A t2–4 cells with the relevant genotype ( MAT a ura3 – 52 leu2 trp1 his7 ade1 – 2 ISE2 top2 - 4 ) were kindly provided by Dr. John L. Nitiss, St. Jude Children's Research Hospital, Memphis TN, USA. This strain and the construct for functional expression of human topoisomerase II α in yeast pMJ1 ( URA3 ) have been described previously [ 66 ]. All yeast strains were transformed with pMJ1 to functionally express human topoisomerase II α in a cell cycle independent fashion. BY4741 wild-type and Δ rad6 , Δ rad50 , Δ rad52 , Δ sae2 and Δ yku70 cells were also transformed with an empty URA3 vector (pYX112). The ICRF-187 and m-AMSA sensitivity of pYX112-transformed cells was assessed to assure that the drug sensitivity of the pMJ1-transformed cells (Table 2 , S1) is related to the ectopic expression of human topoisomerase II α in the cells, which was the case. Transformation and selection was carried out according to standard procedures using lithium acetate cell wall permeabilisation and PEG-mediated DNA uptake by using single-stranded DNA as carrier as described [ 67 ]. Selection was done on SC-URA plates. Three independent pMJ1-transformed yeast clones were selected and propagated for each transformation. All strains were propagated at 30°C, to be subsequently used in clonogenic assays at 34°C. Yeast clonogenic assay The clonogenic sensitivity of the yeast cells towards ICRF-187 and m-AMSA was determined using a clonogenic assay essentially performed as described in [ 7 ]. Briefly, overnight cultures of the strains were grown in SC-URA medium at 34°C at 200 rpm. Cells in log phase were diluted to 2 × 10 6 cells/ml in pre-warmed YPD medium, and 3 ml cultures were exposed to different concentrations of drug at 34°C for 22.5 hours. After drug exposure the samples were diluted up to 10 5 times (depending on the combination of strain and drug used) in distilled sterile water. Yeast cells that were not diluted before plating were spun down by brief centrifugation, and re-suspended in the same volume of sterile water. Next, 200 μl of diluted cells were transferred to SC-URA plates, which were incubated for 5 days at 30°C before counting. 200 to 600 colonies were typically counted for each drug concentration in each single experiment. Finally, the relative survival at the different drug concentrations as compared to the no drug sample was calculated to generate dose-response curves. For each combination of yeast strain and drug, at least three dose-response curves were generated using pMJ1-transformed cells from at least two independent clones (mostly from three). Yeast microarray gene expression analysis Microarray experiments were performed with yeast strain JN362 t2–4 transformed with pMJ1 to functionally express human topoisomerase II α. Fresh colonies were inoculated into YPD medium and grown overnight at 34°C, 180 rpm. The cultures were then diluted to obtain an OD 600 of 0.2. Cultures of 50 ml in YPD medium were first grown for two hours to assure exponential growth of the cells. 1 mg/ml ICRF-187 or 50 μg/ml m-AMSA (equitoxic concentrations) were then added to the cell cultures (a no-drug sample was also included), and the cells were grown for an additional two hours. Each treatment was performed in duplicate. The used concentration of both drugs resulted in a reduction in the clonogenecity of the cells of 50 %. After treatment cells were harvested by centrifugation. Total RNA was isolated by the hot acidic phenol method [ 69 ]. All the steps for cDNA synthesis, cRNA synthesis, biotin labeling and array hybridization to Affymetrix S98 yeast arrays were performed as described in the Affymetrix GeneChip Expression Analysis Technical Manual (Affymetrix), and performed at the microarray core facility at Rigshospitalet, Copenhagen Denmark. Briefly, cDNA was synthesized from 5 μg RNA using a (dT) 24 primer containing a T7 RNA polymerase promoter sequence and SuperScript II reverse transcriptase (Invitrogen) for 1 h at 42°C followed by second-strand synthesis using DNA polymerase I and RNase H digestion followed by isolation of cDNA using GeneChip Sample Cleanup Module (Affymetrix). The cDNA was used as template for synthesis of biotin-labeled cRNA by incubation with biotin-labeled ribonucleotides and T7 RNA polymerase for 5 h at 37°C. Biotin-labeled cRNA was purified using GeneChip Sample Cleanup Module. Biotinylated cRNA was fragmented and 15 μg used for hybridization to Affymetrix Yeast Genome S98 arrays at 45°C for 16 h as described in the Affymetrix users' manual. Washing and array staining with streptavidin-phytoerythrin were performed using the GeneChip Fluidics Station 400 and scanning was performed with a Gene Array Scanner G25 (Agilent technology). Data was analyzed using the DNA-Chip Analyzer (dChip) software [ 70 ]. Real-time PCR analysis The RNA preparations used for microarray analyses were also used for real-time PCR. This analysis was performed on an ABIPrism 7900HT (Applied Biosystems). RNA samples were DNase treated using the DNA-free™ DNase treatment and removal kit (Ambion), and RNA concentrations were measured before conversion to cDNA using the TaqMan RT kit (Applied Bioscience). Priming was performed by random hexamers converting 2 μg RNA pr 100 μl reaction volume, to make 20 ng/μl cDNA. Primers were designed for coding sequences from the Saccharomyces genome database using the Primer 3 input program . All primers were purchased at DNA Technology A/S, with melting temperatures close to 60°C. Reaction mixtures containing the following components at the indicated end-concentrations were prepared. To make a total of 40 μl in sterile water, 20 μl 1x SYBR ® green PCR master mix (Applied Biosystems), 250 nM forward primer, 250 nM reverse primer, and 5 ng template was mixed. Cycling conditions: 95°C for 10 min, followed by 40–45 cycles of 95°C for 15 s and 60°C for 60 s. Relative values of gene expression were calculated with untreated samples as calibrator, and normalized to levels of actin, according to the 2 -ΔΔCt method [ 71 ] and (User bulletin #2, AbiPrism 7700 Sequence Detection System, Applied Biosystems) after primer optimisation and target efficiency evaluation. The following primers were used: HUG1 -forward, AGGCCTTAACCCAAAGCAAT; HUG1 -reverse, TCTTGTTGACACGGTTGCTC; RNR3 -forvard, ATGCATCTCCAGTTCCATCC; RNR3 -reverse, GGGGCAACACTATCTTCCAA; RAD51 -forward, GTGGCGGTGAAGGTAAGTGT; RAD51 -reverse, GTCTAATCCGAACCGCTGAG; RAD54 -forward, CTAAAGCAGGTGGGTGTGGT; RAD54 -reverse, CTTGTTGATCAGCAGCAGGA; ACT1 -forward, CGGTGATGGTGTTACTCACG; ACT1 -reverse, GGCCAAATCGATTCTCAAAA. Mammalian cells The CHO cell lines AA8, irs1SF, CXR3, V3-3, and the hamster lung fibroblast cell line SPD8 were kindly provided by Dr Thomas Helleday, University of Sheffield, UK. AA8 is a wild-type cell line. The AA8-derived irs1SF cell line is XRCC3-defective and has reduced levels of HR [ 37 ]. CXR3 is a human- XRCC3 -cosmid complemented strain of irs1SF, which is proficient in HR [ 37 ]. The V3-3 cell line is DNA-PK cs -deficient and consequently deficient in NHEJ [ 38 ]. The SPD8 cells carry a non-functional hprt gene that can be repaired by HR [ 39 ]. HPRT + cells can then be selected on HAsT medium containing hypoxanthine, L-azaserine and thymidine. When SPD8 cells were not used in the recombination assay they were propagated in medium supplemented with 6-thioguanine to select against spontaneous reversion to the HPRT + phenotype. Human SCLC OC-NYH cells have been described [ 72 ]. Hamster cells were propagated in DMEM medium and OC-NYH cells were propagated in RPMI-1640 medium. All cell culture media were supplemented with 10 % fetal calf serum and 100 U/ml penicillin-streptomycin. Cells were grown in a humidified atmosphere containing 5 % CO 2 in the dark at 37°C. Determination of topoisomerase II activity in crude cell extracts Topoisomerase II activity in crude extract was determined by using a decatenation assay previously described [ 68 ]. Briefly, 200 ng 3 H labeled kDNA isolated from C. fasciculata was incubated with increasing amounts of crude extracts in 20 μl reaction buffer containing 10 mM TRIS-HCl pH 7.7, 50 mM NaCl, 50 mM KCl, 5 mM MgCl 2 , 1 mM EDTA, 15 μg/ml BSA and 1 mM ATP for 20 min at 37°C. After addition of 5x stop buffer (5 % Sarkosyl, 0.0025 % bromophenol blue and 50 % glycerol), unprocessed kDNA network and decatenated DNA circles were separated by filtering, and the amount of unprocessed kDNA in each reaction was determined by scintillation counting. The amount of crude extract required to fully decatenate 200 ng of kDNA under these assay conditions (which is equivalent to 1 U of catalytic activity) was then determined, and the specific activity of the crude extract was calculated as U/μg protein. Mammalian clonogenic assay Four hours prior to continuous treatment with either ICRF-187 or m-AMSA, 250 cells of each of the hamster cell lines were plated onto 100 mm dishes. After 7 days colonies were fixed and strained in methylene blue in methanol (4 mg/ml), and colonies with more than 50 cells were counted. Finally, the relative survival compared to the no drug treatment was calculated, and plotted against drug concentrations to generate dose-response curves. 150 – 250 colonies were typically counted in the "no drug" dishes. Mammalian homologous recombination assay A mammalian recombination assay was performed as described [ 39 ]. Briefly, 1 × 10 6 SPD8 cells were inoculated into 75 cm 2 flasks. When transferred, 6-thioguanine was omitted from the medium. Cells were trypsinised and resuspended in 10 ml medium at 100,000 cells/ml, and exposed to the indicated drug concentrations for 24 hours. To determine clonogenic survival, for each drug-treatment 500 cells were transferred to each of two 100 mm petri dishes containing 10 ml of non-selecting medium and the cells were cultured for 7 days. For selection of recombination events, 300,000 cells were transferred to each of three 100 mm petri dishes containing 10 ml medium supplemented with 50 μM hypoxanthine, 10 μM L-azaserine and 5 μM thymidine and selection was carried out for 10 days. Colonies were fixed by using methylene blue in methanol (4 mg/ml) and counted. Finally, the recombination frequency was determined as the plating efficiency in recombination selective medium divided by the plating efficiency in normal medium, for all concentrations of ICRF-187 or m-AMSA. To enable comparison of recombination frequency at equitoxic levels of m-AMSA and ICRF-187, the recombination frequency was plotted against the relative clonogenic survival of cells receiving only drug. Histone purification Human SCLC OC-NYH cells were grown to sub-confluence and histones were extracted as follows. After the relevant drug treatments, the cells were pelleted and washed in cold PBS, and lysed in lysis buffer (10 mM TRIS-HCl pH = 6.5, 50 mM Sodium Bisulphate, 1% Triton X-100, 10 mM MgCl, 8.6% sucrose) at 4°C by applying 20 strokes in a tight fitting Dounce homogenizer. Released nuclei were pelleted by centrifugation at 2500 g for 10 min at 4°C, and washed in lysis buffer followed by wash buffer (10 mM TRIS-HCl, 13 mM EDTA pH 7.4). The pellet was next resuspended in 100 μl ice-cold 0.4 M H 2 SO 4 , and incubated for 1 hour at 4°C prior to centrifugation. The supernatant was transferred to a clean tube and 1 ml ice-cold acetone was added followed by incubation overnight for histone precipitation. After centrifugation, the pellet was air-dried and resuspended in 40 μl H 2 O, and the protein concentration was determined by Bradford protein assay (Bio-Rad Laboratories) γH2AX western blot Western blotting was performed by loading 10 μg of total histones on a 4–12% gradient gel (NuPageTM Bis-Tris Gel, Invitrogen). Separated proteins were transferred to nitrocellulose membranes (Bio-Rad) which were blocked in 10% skimmed milk (Fluka) and incubated overnight with anti γH2AX primary antibody diluted 1:500 (Upstate Technology, cat no 16–193) followed by detection with goat-anti-mouse (Amersham) 1:2000 for one hour. Detection with ECL Plus™(Amersham) was performed by scanning on STORM™ 840 (Molecular Dynamics Inc), on which the image was optimized and bands quantified by Image Quant™ version 5.0 (Molecular Dynamics). List of abbreviations used BER, Base Excision Repair; CHO, chinese hamster ovary; HR, Homologous Recombination; ICRF-187, (+)-1,2-bis(3,5-dioxopiperazinyl-1-yl)propane; m-AMSA, (N-[4-(9-acridinylamino)-3-methoxyphenyl]methanesulphonanilide); MMR, Mismatch Repair; NER, Nucleotide Excision Repair; NHEJ, Non-Homologous End Joining; PCR, Polymerase Chain Reaction; PRR, Post Replication Repair; SCLC, Small Cell Lung Cancer; SC-URA, Synthetic medium lacking uracil; SSA, Single Strand Anealing; YPD, Medium containing Yeast extract, Peptone and Dextrose. Authors' contributions Lars H. Jensen: Participated in planning the experiments, performed yeast transformations and clonogenic assays, and prepared the manuscript. Marielle Dejligbjerg : Performed γH2AX western blots, primer design, real-time PCR experiments, and data quantitation. Lasse T. Hansen : Performed mammalian clonogenic assays and recombination assays. Morten Grauslund: Performed RNA purification and microarray analysis. Peter B. Jensen : Participated in planning and monitoring the study. Maxwell Sehested: Participated in the initiation and conduction of the study. All authors read and approved the final manuscript. Supplementary Material Additional file 1 Clonogenic sensitivity of mutant single-gene deletion yeast strains towards ICRF-187 and m-AMSA. Clonogenic sensitivity of a panel of human topoisomerase II α-transformed haploid yeast deletion strains towards equitoxic (to wt cells) concentrations of ICRF-187 and m-AMSA. Error-bars represent SEM of 3 – 10 independent experiments. Click here for file Additional file 2 Level of killing of yeast cells used in transcriptional profiling experiments. Fresh colonies of pMJ1-transformed JN362A t2–4 cells were inoculated into YPD medium and grown overnight at 34°C, 150 rpm. The cultures were then diluted into 50 ml YPD medium to obtain an OD 600 of 0.2. After growing the cells for 2 hours to assure exponential growth, equitoxic concentrations of ICRF-187 and m-AMSA were applied and the cells were grown for an additional 2 hours before RNA was isolated. The figure depicts the clonogenecity of drug treated and untreated cells. Exposure of the cells to the two drugs resulted in a reduction of their clonogenecity of approx. 50 %. Error bars-represent SEM of three independent experiments. Click here for file Additional file 3 Transcriptional response towards ICRF-187. A list of yeast genes whose average expression in two independent experiments is induced or repressed more than 1.5 fold by exposure to ICRF-187. Click here for file Additional file 4 Transcriptional response towards m-AMSA. A list of yeast genes whose average expression in two independent experiments is induced or repressed more than 1.5 fold by exposure to m-AMSA. Click here for file Additional file 5 Topoisomerase II activity levels in hamster cell lines. The levels of topoisomerase II catalytic activity in crude extracts from wt and recombination defective hamster cell lines. Error-bars represent SEM of two independent experiments. No difference in the level of topoisomerase II catalytic (DNA strand passage activity) is observed. Click here for file	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC545072.xml
544344	Protein family comparison using statistical models and predicted structural information	Background This paper presents a simple method to increase the sensitivity of protein family comparisons by incorporating secondary structure (SS) information. We build upon the effective information theory approach towards profile-profile comparison described in [Yona & Levitt 2002]. Our method augments profile columns using PSIPRED secondary structure predictions and assesses statistical similarity using information theoretical principles. Results Our tests show that this tool detects more similarities between protein families of distant homology than the previous primary sequence-based method. A very significant improvement in performance is observed when the real secondary structure is used. Conclusions Integration of primary and secondary structure information can substantially improve detection of relationships between remotely related protein families.	Background Detecting an evolutionary relationship between proteins is the basis for functional inference. Existing methods most often rely on sequence information in an attempt to quantify the evolutionary divergence or similarity between the sequences compared. A significant similarity would suggest that the proteins are related. However, in many cases sequences have diverged to the extent that their similarity is undetectable by standard sequence comparison algorithms. Nevertheless, they may still have similar structures and functions [ 1 , 2 ]. It has long been postulated that evolutionary pressure acts upon the three-dimensional structure of proteins and intra-protein interactions rather than at the level of the primary sequence [ 3 , 4 ]. Indeed, there is plenty of evidence to suggest that 3D structure is more conserved than sequence [ 5 , 6 ]. Since the protein structure usually prescribes the function of a protein, relying on structural information (for example, through structure comparison) for functional inference is more effective and reliable than using only the primary sequence. However, although methods of sequencing proteins have become faster and more cost-efficient due to recent technological advancements, methods to determine structure are still in their infancy. In fact, less than 5% of newly sequenced proteins have a known structure. Current empirical processes used to determine structure of proteins are neither efficient nor scalable to use upon the entire known protein space. There have been many attempts to build algorithms that predict protein structure from amino acid sequence. Unfortunately, this is a hard problem, and existing methods are only partially successful [ 7 ]. On the other hand, predicting the secondary structure of a protein has been more successful. There are various algorithms that predict the secondary structure from primary amino acid sequence information alone [ 8 - 13 ]. The accuracies of these algorithms have been steadily increasing, and one of the most successful algorithms to date is PSIPRED [ 13 ], which has an average accuracy of about 80%. Since the architecture of the secondary structure elements of a protein affects its global structure, it has been suggested that secondary structure information can be used to detect subtle similarities between proteins that have diverged substantially in the course of evolution. This principle was tested in [ 14 ] where a dynamic programming algorithm with a secondary-structure based scoring matrix was used to compare protein sequences over the alphabet of secondary structures. However, relying solely on secondary structure information might lead to poor performance overall, as much of the original information about the individual amino acids is lost. Alternatively, one can use both representations to assess protein similarity. Incorporating secondary structure information into protein comparison is not a new idea. Several researchers have attempted to boost performance and sensitivity of various methods by adding this extra degree of information with some success. Yu et al. encoded functionally conserved sequence patterns into probabilistic structural models (that comprise a family of hidden Markov models) [ 15 ]. The models were benchmarked against the trypsin-like serine protease family and the globin family, and in both cases proved to have high specificity and sensitivity compared to the models in use at the time (primarily, BLAST) in remote homology detection. One of the limitations of this model, however, was the reliance on threading methods requiring at least one determined structure to build a model. Hedman et al. [ 16 ] included information about predicted transmembrane segments into the standard Smith-Waterman and profile-search algorithms for membrane proteins by adding an extra delta (score) when two residues that are both predicted to belong to transmembrane segments are aligned. This method was found to improve the detection rate, mainly by increasing specificity (ie. decreasing the number of false positives). Ginalski et al. [ 17 ] generalized a method of creating "meta profiles" by combining sequence information with predicted secondary structure information. Total scores were calculated by summing the raw score obtained from the shifted dot product of the sequence profile vectors and the shifted dot product of the secondary structure probability vectors (weighted by some factor). This technique was derived from hybrid threading approaches and was found to be more sensitive than the sequence-only approach or sequence-to-structure threading approach. Teodorescu et al. [ 18 ] proposed a linear combination of threading and sequence-alignment to produce a single (mixed) scoring table. This method was found to be particularly sensitive in detecting sequences with less than 25% of sequence identity, yet with similar structures. The final model outperforms the individual scoring methods. These and similar studies have indicated that the incorporation of secondary structure information, even if predicted, can increase sensitivity and specificity of a protein comparison model. Here we describe a method that integrates secondary structure information with primary sequence information in a single scoring scheme, using a single statistical representation. The model can be applied to any protein family and does not require the application of expensive threading algorithms. Our method extends our previous work on profile-profile comparison [ 19 ]. Specifically, we use the profile representation (generated by PSI-BLAST) as a statistical model of a protein family and augment the profile with structural information. We then compare profiles of different protein families, in search of possible remote kinship, using an information theory-based scoring function. By comparing models of protein families we can detect similarities that are not detected when comparing individual sequences. We show that the new algorithm improves performance and can detect more similarities between remote protein families. These similarities can be used to classify protein families into super-families and detect higher order structure within the protein space. Methods and Results Data sets We use a data set of domain families derived from the SCOP classification of protein structures [ 20 ], release 1.50. This set contains 23,780 protein domains classified into 1,287 protein families. Each of the 1,287 families is represented by a profile that was generated using PSI-BLAST [ 21 ]. The seed of the profile was selected to be the sequence whose average distance from all other members of the family is the smallest. Families for which there is only one member, or for which PSI-BLAST failed to generate a profile, were represented by a profile generated directly from the seed sequence by using probabilities derived from the original BLOSUM62 frequency matrix [ 22 ]. A subset of 456 families was used in our study, all of which belong to superfamilies that contain at least 3 families. A smaller subset of 120 families was used for parameter optimization. Sequence profiles The PSI-BLAST profiles are the basis for our representation of a protein family. Each profile is a n -tuple of probability distributions of amino acids, derived from a group of related proteins, where n is the length of the multiple alignment of these proteins. It is represented in software as a two dimensional matrix of 20 rows and n columns, where each column (known as a profile column), is a probability distribution p over the 20 amino acids in one position in the multiple alignment. These profile columns form the basis of profile-profile comparisons. Secondary structure information We use two types of secondary structure information in our experiments: true information and predicted information. The true secondary structure information is gleaned from the PDB files of the seed proteins using STRIDE [ 23 ]. Stride defines eight types of secondary structures b , B , C , E , H , I , G , T where b and B stand for Bridge, C = Coil, E = Strand, H = AlphaHelix, I = PiHelix, G = 310Helix and T = Turn. We reduce this set to the three main secondary structures (helix, strand and coil) by mapping H, I, G to H, and b, B, C, T to C. The predicted secondary structure information is predicted using PSIPRED [ 13 ]. PSIPRED uses the intermediate sequence profiles generated by PSI-BLAST as input for the prediction algorithm. This profile matrix is fed into a standard feed-forward back-propagation neural network with a single hidden layer using a window of 15 residues. This net has three output units corresponding to each of the three states of secondary structures. Another window of 15 positions of these three outputs (per amino acid) are then used as input into a second neural network to filter and smooth outputs from the main network. The final output is the probability that a certain amino acid position in the seed sequence of a profile is in a coil, helix, or strand. PSIPRED reports an average Q3 score of approximately 80% accuracy. Integrating secondary structure with primary structure Apriori, it is unclear how one should integrate secondary structure with primary structure in a single model. For example, one might think of a representation over a generalized alphabet, that considers all possible pair combinations of amino acids and secondary structure elements. Assuming independence between positions (which is the underlying assumption of position specific scoring matrices, as well as of HMMs that are used in computational biology), then this representation implies that for each position i we have a statistical source that emits amino acid a and secondary structure s with probability P i ( a , s ) such that and every position can be represented by a vector of 60 probabilities over this pair alphabet. This representation implicitly implies that the amino acid emitted and the secondary structure are two different features of the objects generated by the source, while in reality the secondary structure is not a "character" or an independent property of the emitted objects, but rather a characteristic of the source itself that is usually unknown. This property introduces some special constraints on the distribution of amino acids that are emitted by the source. In other words, the secondary structure and the amino acid distribution in a position are strongly dependent on each other, but one is hidden while the other is visible . Noting that P i ( a , s ) can be written as P i ( a / s ) P i ( s ), we can decompose the parameter space into the parameters of the secondary structure distribution, and the parameters of the conditional probability distributions over amino acids. However, the typical amino acid distributions that are available from multiple alignments of protein families differ from these conditional probability distributions by definition. Furthermore, there are other subtleties that one should bear in mind when designing an integrated statistical model for a protein family. More precisely, assume we have a protein family, where all proteins adopt a certain structural conformation of length n . This conformation can be described in terms of the set of 3D coordinates of the n positions, or in terms of the set of distances between coordinates S = ( ) where is the set of distances from the i -th residue to all other residues – the latter being more amenable for a representation as a statistical source, as it is invariant to translations and rotations. Although there is structural variation across the different instances of the protein family, it is significantly smaller than the sequence variation, and we will assume that a single consensus conformation S reliably describes the protein family. The structural conformation determines the statistical properties of the source distributions. Namely, it induces certain constraints on the sequence space that can be mapped to that conformation, based on the physical properties of its topology. In other words, it induces a probability distribution over the sequence space of O (20 n ) sequences that can be mapped to that conformation P ( a 1 , a 2 , ..., a n / S ). Note that due to convergent evolution it is possible that two disconnected regions in the sequence space (two families of homologous proteins) will be mapped to the same conformation (although experimental evidence and simulation results [ 24 ] suggest that this is not very likely, and for most protein families it is reasonable to assume that the sequence space that is mapped to a structural conformation is connected). This 20 n -dimensional distribution clearly introduces dependencies between remote positions, and the exact probability distribution in a position depends on the amino acids observed in all other positions P ( a i / a 1 , a 2 , ..., a i -1 , a i + 1 , ..., a n , S ). Accurate knowledge of the all-position probability distribution P ( a 1 , a 2 , ..., a n / S ) would allow one to compare two sources of protein families theoretically by comparing these high-dimensional distributions. However, because of (limited) data availability and for mathematical simplicity, the marginal probabilities are usually used in practice to describe the source. Given a multiple alignment of a specific protein family, and the corresponding profile, the empirical distribution of amino acids at position i , denoted by , is essentially the marginal probability of amino acids at that position, as constrained by the global conformation , i.e. The complete model is represented as a set of marginal probability distributions, one per position. So far we have not considered the secondary structures explicitly. The secondary structure sequence s is a reduced representation of S that, while incomplete, describes quite accurately the topology of the protein. Given S , the knowledge of s however does not affect the distribution of amino acids at a position, i.e. P i ( a / s , S ) = P i ( a / S ) Nevertheless, the secondary structure information can still be useful when comparing protein families. This is because some information is lost if one is to use just the marginal amino acid distributions. For example, the same marginal amino acid distribution can be observed in different secondary structure conformations and on the other hand, even highly similar fragments of secondary structures can be associated with different amino acid distributions. The most effective way of comparing two protein families is by comparing their (consensus) structural conformations S 1 and S 2 . Indeed, it has been shown that structure comparison is much more effective in detecting remotely related families [ 19 , 20 , 25 ]. In statistical terms, one can formulate the problem of comparing consensus structures S 1 and S 2 as comparing two sources that induce different probability distributions over the conformation space P 1 ( S ) and P 2 ( S ). However, characterizing these distributions is very difficult. Moreover, convergent evolution might result in two different sequence sources with structurally similar conformations. These relations are usually perceived weaker than families that are similar both in sequence and structure [ 20 ]. Therefore, a proper comparison should account for both the primary and tertiary structure. In statistical terms, we are interested in comparing the joint distributions and , where the distributions are again marginalized over all positions other than i , and is a vector of inter-residue distances. The joint distributions can be rewritten as where the last step uses the more accurate marginal probabilities P i ( a / S ) that are based on all vectors of inter-residue distances (and match the empirical distributions ). As was mentioned earlier, obtaining S is difficult (and therefore also characterizing the distributions of inter-residue distances). On the other hand, secondary structure (which can be viewed as an approximation of S ) is more readily available, and can be predicted quite reliably from sequence information. Therefore we suggest to approximate where P i ( s ) is the probability to observe a secondary structure s at the i -th position. (When the secondary structure is known the distribution over secondary structures assigns probability 1 to one of the structures and zero otherwise. However, with predicted information, each state is usually assigned a non-zero probability based on the amino acids in that position and neighboring positions.) Plugging in the empirical distributions for P i ( a / S ) we get i.e., the empirical distribution of amino acids at a position, , is conditionally independent of the distribution P i ( s ). Therefore, to completely describe the source one needs to provide the parameters of the marginal distribution of amino acids, and the parameters of the secondary structure distribution. Since the two distributions are assumed independent, they are amenable to a representation in which their parameters are appended together. I.e. the secondary structure probabilities are appended to the probabilities of the 20 amino acids. Our method is based on an extension of the original profile representation in [ 19 ]. Using the three PSIPRED probabilities, we augment the profile columns of primary information to make a probability distribution over 23 values (the 20 amino acids plus 3 secondary structures). Note that by doing so, each profile column is now dependent upon and contains information about its neighbors, since PSIPRED uses the profile columns surrounding each profile column to deduce the probability that the position in question is in a specific secondary structural conformation. This is the key element that enhances the accuracy of this tool in protein family comparisons. Moreover, the method is "self-contained" in the sense that for the secondary structure prediction, PSIPRED uses the same profiles that are generated by PSI-BLAST. To use the profile-profile metrics described next, the 23-dimensional profile columns have to be normalized to conform with probability distributions. However, apriori it is not clear if the primary information and the secondary structure information should be weighted equally. To control the impact of the secondary structure information on the representation we introduce a mixing parameter γ that ranges from 0 to 1. The secondary structure probabilities are normalized such that they sum to γ while the amino acid probabilities are normalized such that they sum to 1 - γ . The higher γ is, the more dependent the profile column is upon secondary structure information. This parameter is optimized as described in section 'Parameter optimization'. Note that our normalization maintains the conditional independence of the two types (primary and secondary), as described above. Each component of the extended profile can be viewed as a sub-profile. Since each one of the two components is summed independently to a non-zero probability then two symbols must be "emitted": an amino acid and a secondary structure. Profile-Profile comparison In this section we review the main elements of our profile-profile comparison algorithm that was introduced in [ 19 ]. We compare profiles using the dynamic programming algorithm with an information theoretic-based scoring function to score pairs of profile columns. Given two profiles P = p 1 p 2 p 3 ... p n and Q = q 1 q 2 q 3 ... q m , where n and m are the lengths of the profiles (the number of positions or columns) and p i , q j are probability distributions over the 23 letter alphabet of amino acids and secondary structures, we define the similarity score between two columns p i and q j based on their statistical similarity. The similarity score is composed of two elements: the divergence score and the significance score. The divergence score To estimate the divergence of two probability distributions we use the Jensen-Shannon (JS) divergence measure [ 26 ]. Given two (empirical) probability distributions p and q , for every 0 ≤ λ ≤ 1, the λ -JS divergence is defined as where D KL [ p \|\| q ] is the Kullback-Leibler (KL) divergence [ 27 ], defined as and r = λ p + (1 - λ ) q can be considered as the most likely common source distribution of both distributions p and q , with λ as a prior weight (here set to 0.5). We call the corresponding measure the divergence score and denote it by D JS . This measure is symmetric and ranges between 0 and 1, where the divergence for identical distributions is 0. Besides being symmetric and bounded, an attractive feature of the D JS divergence measure is that it is proportional to the minus logarithm of the probability that the two empirical distributions represent samples drawn from the same ("common") source distribution [ 28 ]. It has also been shown that is a metric [ 29 ]. The significance score The divergence score measures one aspect of the statistical similarity of p and q : their relative distance. However, it does not consider the uniqueness of the two distributions. A match between two distributions that resemble the background distribution is not as significant as a match of two distributions that resemble each other, but are very different from the background distribution. In other words, the more unique the distributions are (and hence, also their common source), the more significant is a match between them. To assess the significance score S of a match we measure the JS divergence of the (common) source distribution, r , from the base (background) distribution P 0 . S = D JS [ r \|\| P 0 ] In this study the background distribution is composed of two components: the background distribution of amino acids (estimated from a large sequence database) and the background distribution of secondary structure elements (estimated from all PDB structures). The components are mixed using the same mixing parameter γ described in section 'Integrating secondary structure with primary structure'. The significance measure reflects the probability that the source distribution, r , could have been obtained by chance. The higher r is, the more distinctive the common source distribution, and the lower the probability that it could have been obtained by chance. The similarity score We define the similarity score of two probability distributions p and q as a combination of the divergence score and the significance score: With this expression, the similarity score of two similar distributions ( D → 0) whose common source is far from the background distribution ( S → 1), tends to one. On the other hand, the similarity score of two dissimilar distributions ( D → 1) whose most likely common source distribution resembles the background distribution ( S → 0) tends to zero. This scoring scheme also distinguishes two distributions that each are similar to the background distribution ( D → 0 and S → 0 giving Score - 1/2) from two dissimilar distributions, but whose common source is similar to the background distribution ( D → 1 and S → 0 giving Score = 0). In a recent study [ 30 ] it has been shown that this scoring function is one of the best, when compared to other methods for profile-profile comparison. Note that our measures are functionals of the probability distributions, based on variations of the entropy function, and specifically the KL divergence function. One of the nice properties of this function is that it is additive in the following sense. Assume we have a probability distribution p over a set X that is obtained by "mixing" two probability distributions over two disjoint subsets: p 1 over the subset X 1 and p 2 over the subset X 2 (where X = X 1 ∪ X 2 and X 1 ∩ X 2 = θ ). Let γ be the mixing parameter, i.e. the total weight of the first distribution p 1 in the combined distribution p . Assume q is obtained in a similar manner from q 1 and q 2 . Then, In other words, this measure preserves independence between the two subsets. Therefore, with our extended profile representation, the new functionals are simply a weighted sum of the individual functionals over the subsets X 1 (the secondary structure) and X 2 (the primary structure). Note however that this property holds for the divergence and the significance measures but not for the final similarity score that is a combination of the divergence and the significance scores. An alternative would be to compute the divergence, significance and similarity scores independently for the secondary and primary structures, and then combine the two similarity scores into one, with weights γ and (1 - γ ) respectively. The effect of secondary structure on the pairwise scores It is interesting to compare the similarity scores before and after the addition of secondary structure information. To assess the impact of this information, we computed the distribution of similarity scores for five types of profile columns, depending on the type of their seed amino acid. We refer to the amino acid at position i of the seed sequence (see section 'Data sets') as the seed amino acid of the i -th profile column. Two seed amino acids are defined as similar, neutral, or dissimilar based on their BLOSUM62 scoring matrix [ 22 ], with positive, zero and negative substitution scores respectively. The five types of column pairs are: (1) a column with itself ( identical columns ), (2) different columns that are associated with the same seed amino acid ( strongly similar columns ) (3) different columns that are associated with similar seed amino acids ( similar columns ), (4) different columns with mutually neutral seed amino acids ( neutral columns ), and (5) different columns with dissimilar seed amino acids ( dissimilar columns ). We repeated this calculation before and after the integration of true secondary structure information (using the optimal mixing parameter γ , see section 'Parameter optimization') and the results are plotted in Figure 1 . As the figure indicates, there is a slight shift between the distributions, and the addition of secondary structure information pushes the distributions further apart, decreasing the distribution overlap, as desired. Although the differences are small (due to the very small value of the optimal mixing parameter), the effect on the performance is significant as is demonstrated in section 'Discussion'. Comparison of scoring functions We compared our information-theoretic scores to other popular scoring schemes. We tested the correlation scores based on the scalar product of the vectors (as was suggested in [ 31 ]). We also tested the ALLR (Average Log Likelihood Ratio) scoring function that was suggested in [ 32 ]. This scoring function is also based on information-theoretic principles, and resembles ours. Given two empirical probability distributions p and q , their ALLR score is defined as where n p ( n q ) is the number of total counts from which p ( q ) is derived, and P0 is the background distribution. We computed the correlation scores and ALLR scores for the same sets of columns defined in the previous section and compared it to the information-theoretic scores (Figure 2 ). Note that the correlation scores are less successful in distinguishing related columns from columns which are likely to be unrelated (compare Figure 2a and Figure 2b ). The overlap is larger and the tail of the fifth distribution (dissimilar columns) falls well within the first distribution (identical columns). Specifically, 24% of the pairs of dissimilar columns have correlation scores that overlap with scores of identical columns, compared to only 2.1% when using our similarity scores. We believe that this may affect the performance significantly. On the other hand, the ALLR scores have very similar properties to ours, although the overlap between dissimilar columns and identical columns is greater (4.4%). Parameter optimization Our algorithm ( prof _ ss ) depends on several parameters: (1) a shift parameter is introduced to convert the similarity scores to scores that are suitable for local protein comparison (other transformations were tested in [ 19 ] and proven less effective); (2) gap penalties for the dynamic programming algorithm; (3) the mixing parameter γ Shift parameter and gap penalties as Figure 2a shows, the distributions of identical columns (red line) and distributions with dissimilar seed amino acids (black line) are quite well separated around 0.5. In addition, distributions with mutually neutral seed amino acids peak at a similarity score around 0.45. Note that the new similarity scores (after the addition of the secondary structure information) preserve the overall behavior (quantitatively and qualitatively) as the old similarity scores (see Figure 1 ). The mean of the scores is unchanged and only the variance has increased. Therefore, we decided to maintain the same set of parameters that were optimized in [ 19 ]. Specifically, we used the same shift value of 0.45 and the same gap penalties of 2 (gap opening) and 0.2 (gap extension). We have also tested position-specific gap penalties based on the SS information, but without any apparent improvement in performance. Mixing parameter To estimate the best value for γ we used a subset of 120 families and assessed the performance for different values of γ . Our performance evaluation procedure works as follows: true relationships are defined to be between those families that share a superfamily and all others are defined as false relationships. For each family within the test set, we calculate the profile-profile similarity against all 1287 families for a single value of the mixing parameter γ . These results are sorted by raw score and the number of true family-family relationships are counted before the first false relationship is detected (this is basically a sum of ROC1 scores). The tradeoff between the primary sequence information and secondary structure probabilities was varied from zero to one. With zero dependence on secondary structure the method is equivalent to prof _ sim (profile-profile comparison based on just primary structure). The results are shown in Figure 3 . As the graph indicates, setting γ = 0.055 (i.e. 0.055 weight on the secondary structure information and 0.945 on the sequence information) gave the best performance. (Note that if each secondary structure was given as much weight as a single amino acid γ would be or ~0.13). When only secondary structure information was used ( γ = 1), the performance was much worse than when only sequence information was used ( γ = 0). These corner-case results and the fact that the best results were obtained with γ << 0.5 suggest that for protein family comparison, the coarse-grained secondary structure information is noisier and less reliable than sequence information. However, as the graphs indicate, using both sources of information clearly improves performance. Our tests were done using actual secondary structure information in the profile; however, similar results were obtained when the predicted information was used for one or both of the profiles (see Figure 3b ). Statistical significance To differentiate true similarity values from those that may be observed by chance, it is essential to establish a baseline empirical distribution for the scores. Here we used the statistical framework of the extreme value distribution (EVD). Although rigorous mathematical proof has not been found for local gapped similarity scores, empirical studies have shown that the distribution of these scores can be approximated by this distribution. An empirically fit EVD also has the benefits of being a true fit to the quirks of a particular protein family. Three such distributions were established to assess the significance of the profile-profile matches. All distributions were fit with the 'fit' function in gnuplot using the nonlinear least-squares (NLLS) Marquardt-Levenberg algorithm. The first distribution is based upon comparisons between unrelated families (defined as families that belong to different SCOP classes and do not share significant structural similarity). This distribution is useful in that it can be used to assess the significance of a score in comparing any pair of protein families, without further need for computations. Practically, this aggregates all comparisons between non-related families into a single list. This is essentially the distribution of similarity scores of random profiles, as shown in Figure 4a . By fitting an EVD to this distribution we can estimate the statistical significance (e-value) of any raw similarity score. We refer to this method as the uniform approach (uniform parameters). The second distribution is similar to the first, except a correction was made for the length of a profile, similar to the approach employed by FASTA [ 33 ]. By chance, the raw score of a profile-profile comparison is greater for those profiles with many more residues than the score of two smaller profiles. To correct for this occurrence, all raw scores were fit to a logarithmic curve of the product of the two profile lengths. The mean and variance of this fit was used to calculate a zscore. Accounting for undersampling at the ends of the spectrum, the means were fit to a linear curve and the variance was constant throughout. The distribution of zscores was then fit to an EVD, as is shown in Figure 4b . This distribution estimates better the statistical significance of raw similarity scores since it accounts for the biases introduced due to the lengths of the profiles. The third distribution proves to be the best approach in assessing significance of matches with a particular profile. This distribution is created on a per-family basis. The scores of each family against all (unrelated) SCOP families were fit to an EVD. Since many of the family profiles are unrelated to the query family, the corresponding scores provide a relatively reliable baseline distribution. This approach is a robust method to assessing the significance of matches for a particular profile since it allows for any unusual properties of the query profile (like unusual amino acid composition) and the parameters are adjusted accordingly (see Figure 5 ). Once again, from the fitted EVD, the e-value of the raw similarity scores is estimated from this fit. The third method of measuring statistical significance is self-calibrating and provably more accurate than the previous two methods, and our performance evaluation tests indicate that this is the best method overall (see Figure 6 ). However, it is an intractable method when given a single pair of profiles to compare, since there is no prior knowledge about the baseline distributions of either profile. As a result, we must rely on the second method to measure statistical significance in these cases. Discussion We evaluate the performance of our algorithm using the SCOP database as a benchmark and two measures of performance. The first counts the number of weak relationships between protein families (as implied by the SCOP classification) that can be detected with our method. Specifically, each family in our test set is compared with all other protein families and the results are sorted based on the p-value. Given the sorted list we count the number of true family-family relationships that are detected before the first false positive occurs. This measure is applied to each family individually , and the results, summed over all families in the test set are given in Table 1 . We compare our results to Gapped-BLAST, PSI-BLAST and prof _ sim (as reported in [ 19 ]). Usually a false positive is defined as a relation between families that do not belong to the same superfamily. This popular criterion, however, is somewhat strict as relations between families that belong to the same fold can also be considered as positives. We use the following terminology to distinguish between the different types of "false positives". We define a relationship between two protein families to be a true relationship if both families belong to the same superfamily, a possible relationship if both families belong to the same SCOP fold, a weak relationship if they belong to the same class, suspicious if they belong to different classes (excluding the case of an all-alpha ↔ all-beta pair) and an error if one family is all-alpha and the second is all-beta. We repeat the procedure described above, each time using a different definition of a false positive. The results are summarized in Table 1 . The second measure we use is the receiver operator characteristic (ROC) measure, a common measure in assessing sensitivity and selectivity. Given a sorted list of results, the ROC index measures the area under the curve that plots the positives versus the negatives. Maximal performance translates to a perfect separation and a maximal normalized ROC score of 1. The ROC-N measure is a variation over the ROC measure, where the plot is truncated at N negatives. In other words, the ROC-N measure is the number of true positives detected up to N false positives. Here we used the popular ROC-50 measure. To obtain the ROC-50 scores for each method we pool together all pairwise comparisons for all protein families, and sort them by their normalized e-value. The number of true positives is aggregated until 50 false positives occur. As before, we repeated this procedure with different definitions of false positives, and the results are summarized in Table 2 . A detailed break-up of the pairwise similarities detected with each method is given in Table 3 (using the most strict definition of a false positive). Note that prof _ ss improves over prof _ sim (for all types of false positives) although the improvement is smaller compared to the one reported in Table 1 . The difference in performance is striking when the true secondary structure information is used. Despite the moderate contribution to the profile (the optimal γ was set to 0.055), the new algorithm almost doubles the number of pairwise relationships that are detected. Examples In this section we give several interesting examples of alignments between remote protein families that exemplify the differences between sequence-based profile-profile alignments and the new generalized profile alignments. The "winged helix" DNA-binding domain superfamily This superfamily is part of the DNA/RNA-binding 3-helical bundle fold. We compared two families from that superfamily: the restriction endonuclease FokI, N-terminal recognition domain (family a.4.5.12, seed scop domain d2foka3), and the replication terminator protein (family a.4.5.7, seed scop domain dlbm9a_). Although designated as all-alpha, proteins in this superfamily contain a small beta-sheet at the core. The similar substructures have three alpha helices and a couple beta strands, prof _ sim is able to roughly match up the helices but not the beta strands with a rms of 11.96. The predicted secondary structure does not improve the alignment in this case, however, when the true secondary structure is used, prof _ ss is able to completely align the helices as well as most of the strands with a much better rms of 4.45 (Figure 7 and Figure 8 ). The concanavalin A-like lectins/glucanases superfamily This superfamily belongs to the concanavalin A-like lectins/glucanases fold, characterized by a sandwich structure with 12–14 strands in 2 sheets. We compared two families in this superfamily: the beta-Glucanase-like family (b.29.1.2, seed domain dlcpm__) and the vibrio cholerae sialidase, N-terminal and insertion domains (b.29.1.8, seed domain dlkit_2). These class beta proteins have complex topology and are hard to align even with structure alignment algorithms. In this example, the two sets of beta sheets are nicely aligned by prof _ ss both when using the predicted information and the true secondary structure information. On the other hand, prof _ sim is unable to align the sheets at all (see Figure 9 and Figure 10 ). The alpha/beta-Hydrolases superfamily The alpha/beta-Hydrolases belong to the fold by the same name. Proteins with that fold are composed of 3 layers at the core, of alpha/beta/alpha. We compared two families in this superfamily: the carbon-carbon bond hydrolase family (c.69.1.10, seed domain dlc4xa_) and the bromoperoxidase A2 family (c.69.1.12, seed domain dlbrt__). These are large and complex proteins with many helices and strands. prof _ sim reports an alignment that aligns perfectly one small alpha helix and two beta strands. With predicted secondary structures, prof _ ss is able to generate a much longer alignment, with γ alpha helices and 4 beta strands. The alignment is not perfectly in sync, but all secondary structures are roughly in position. When using the true secondary structure information in prof _ ss the alignment improves and a better overlap is observed (see Figure 11 and Figure 12 ). Conclusion This paper presents a simple method to improve remote homology detection between protein families. We use statistical models of protein families in the form of profiles, and by incorporating secondary structure information within that model, we can reuse existing comparison methods for comparing profiles. It is shown that this method improves over the previous method that is based only on primary sequence information. As opposed to other methods that compare single proteins, our method compares models of protein families. Instead of summing over different models, our model combines structural and primary sequence information within the profile itself. Our method allows us to explore a wide range of scenarios, between purely sequence-based representation and a purely secondary-structure based representation. The optimization of the single mixing parameter shows that the slight incorporation of predicted secondary structural information is invaluable. Since predicted structure information in PSIPRED comes from neighboring profile columns, this proves that each profile column confers extra information that is relevant to its neighbors and is useful to inferring protein relationships. Furthermore, it is shown that if true secondary structure information is used, performance improvements are very significant and the number of relationships that can be detected is almost doubled. We conclude that despite the high overall accuracy of the secondary structure prediction method, its imperfect nature can greatly affect the performance. However, our method can be generalized to any secondary structure prediction method that produces estimated probabilities for secondary structure, so should a new prediction method be found that performs better than the current methods, the model presented here is expected to reflect the improved performance and consequently improve homology detection. Authors' contributions RC extended the prof _ sim program and integrated secondary structure information, optimized the model, ran experiments, and analyzed the result sets. GY conceived of the study, designed the model and analyzed the results.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC544344.xml
529433	The Quest for a Vaccine That Yields Tumor-Killing T cells	null	The immune system has a remarkable capacity for fending off infectious diseases, and it has become clear that these same defenses can recognize and destroy cancer cells. In fact, they do so on an ongoing basis, and cancer develops only when immune surveillance breaks down. Many patients with established tumors also mount an immune response against some antigens that are specific to, or enriched in, the tumor. This response, however, is rarely effective against the disease. The idea of enlisting the immune system to fight cancer has been around for a long time, and has led to the development of various cancer vaccines designed to alert the immune system to the presence of a tumor and to induce a response that, selectively and potently, will eliminate tumor cells. Vaccines include whole tumor extracts or specific proteins and peptides that are selectively expressed or enriched in tumors, by themselves or with a variety of adjuvants. There have been some spectacular successes, in particular with immune therapy to malignant melanoma, a tumor type that seems naturally to be more immunogenic than others. However, even in melanoma, success is usually restricted to a fraction of the patients, with no obvious explanation of why the strategy works for a particular patient and fails in most others. The emphasis has consequently shifted from clinical outcomes to monitoring a patient's immune response. What type of response is necessary and sufficient to eliminate tumor cells is still unclear, but the hope is that understanding the immune response in patients that show clinical benefit will answer that question. Peter Lee and colleagues used state-of-the art technology to dissect the endogenous immune response to vaccination with heteroclitic melanoma peptides, i.e., melanoma-associated peptides that have been engineered to elicit a stronger immune response. They focused on cytotoxic T lymphocytes (CTLs), and compared CTL clones from four melanoma patients who had vaccine-induced T cell responses and two melanoma patients with spontaneous anti-tumor T cell responses. The researchers analyzed several hundred CTL clones (to get a sense for the complexity of the responses in individual patients) for T cell receptor variable chain beta expression, recognition efficiency, and ability to lyse target melanoma cells. Most T cells isolated from vaccinated patients were poor at tumor cell lysis compared with T cells from endogenous responses to cancer. Melanoma—a prime target of cancer vaccines (Photo: Timothy Triche, National Cancer Institute) The authors suggest that the high doses of peptides administered in vaccinations and the increased binding capacity of heteroclitic peptides to major histocompatibility complex molecules—the very quality that makes them more immunogenic—induce many T cells with low recognition efficiency for the native peptides they encounter on the tumor cells. Their findings also bring into question the ability to deduce the recognition efficiency and tumor reactivity of T cell responses from ELISPOT and tetramer staining assays—the two standard measures of T cell responses to vaccines—which has implications for rational vaccine design in general.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC529433.xml
509244	8-Cl-Adenosine enhances 1,25-dihydroxyvitamin D3-induced growth inhibition without affecting 1,25-dihydroxyvitamin D3-stimulated differentiation of primary mouse epidermal keratinocytes	Background Epidermal keratinocytes continuously proliferate and differentiate to form the mechanical and water permeability barrier that makes terrestrial life possible. In certain skin diseases, these processes become dysregulated, resulting in abnormal barrier formation. In particular, skin diseases such as psoriasis, actinic keratosis and basal and squamous cell carcinomas are characterized by hyperproliferation and aberrant or absent differentiation of epidermal keratinocytes. We previously demonstrated that 8-Cl-adenosine (8-Cl-Ado) can induce keratinocyte growth arrest without inducing differentiation. Results To determine if this agent might be useful in treating hyperproliferative skin disorders, we investigated whether 8-Cl-Ado could enhance the ability of 1,25-dihydroxyvitamin D 3 [1,25(OH) 2 D 3 ], a known keratinocyte differentiating agent and a clinical treatment for psoriasis, to inhibit keratinocyte growth. We found that low concentrations of 8-Cl-Ado and 1,25(OH) 2 D 3 appeared to act additively to reduce proliferation of primary mouse epidermal keratinocytes. However, another agent (transforming growth factor-beta) that triggers growth arrest without inducing differentiation also coincidentally inhibits differentiation elicited by other agents; inhibition of differentiation is suboptimal for treating skin disorders, as differentiation is often already reduced. Thus, we determined whether 8-Cl-Ado also decreased keratinocyte differentiation induced by 1,25(OH) 2 D 3 , as measured using the early and late differentiation markers, keratin 1 protein levels and transglutaminase activity, respectively. 8-Cl-Ado did not affect 1,25(OH) 2 D 3 -stimulated keratin 1 protein expression or transglutaminase activity. Conclusions Our results suggest that 8-Cl-Ado might be useful in combination with differentiating agents for the treatment of hyperproliferative disorders of the skin.	Background The epidermis of the skin serves as a mechanical and water permeability barrier essential for terrestrial life (reviewed in [ 1 ]) and is composed primarily of epidermal keratinocytes. These keratinocytes stratify to form several layers. The deepest layer, the stratum basalis or basal layer comprises proliferating cells that continuously divide to regenerate cells lost to the environment. As the cells migrate upward into the first differentiated layer, the stratum spinousum or spinous layer, they cease proliferating and begin to express the intermediate filament proteins, the mature keratins 1 and 10. This early differentiation is followed by a late differentiation program in the stratum granulosum or granular layer, which is marked by the expression of other structural proteins, such as filaggrin and loricrin, and by increased activity of the enzyme, transglutaminase, which forms highly durable γ-glutamyl-ε-lysyl bonds to cross-link the proteins into a tough and resistant shell underneath the plasma membrane. At the boundary of the granular layer and the outermost stratum corneum, or cornified layer, the keratinocytes terminally differentiate, degrading their nuclei and other organelles and releasing lamellar bodies, the lipid contents of which form a water-impermeant barrier. The squames, the flattened remnants of the keratinocytes, and the lipids from the lamellar bodies form a sort of brick and mortar, to prevent water loss, microbial invasion and/or other mechanical insults (reviewed in [ 2 - 4 ]). 1,25-Dihydroxyvitamin D 3 [1,25(OH) 2 D 3 ] is a known regulator of this process of keratinocyte growth and differentiation (reviewed in [ 2 , 5 ]). In vitro , 1,25(OH) 2 D 3 inhibits keratinocyte proliferation and stimulates the expression of numerous keratinocyte differentiation markers (reviewed in [ 2 , 6 ]). In vivo a physiologic role for 1,25(OH) 2 D 3 in regulating keratinocyte differentiation is suggested by several lines of evidence: (1) keratinocytes express both the 25-hydroxylase and the 1α-hydroxylase which converts inactive vitamin D 3 to its active 1,25-dihydroxy metabolite (reviewed in [ 2 , 6 ]); (2) receptors for 1,25(OH) 2 D 3 are present in the skin and in epidermal keratinocytes in vitro [ 7 - 11 ]; and (3) Vitamin D receptor null mice exhibit altered skin function, characterized by abnormal hair follicles and reduced expression of several keratinocyte differentiation markers [ 12 ]. Furthermore, 1,25(OH) 2 D 3 and its structural analogs have been used as effective treatments for psoriasis, a human skin disease characterized by inflammation and by hyperproliferation and abnormal differentiation of keratinocytes (reviewed in [ 13 , 14 ]). 8-Chloro-cyclic-adenosine monophosphate (8-Cl-cAMP) is known to inhibit growth and to induce apoptosis in a variety of cancer cells [ 15 - 18 ], suggesting its potential utility as an anti-cancer drug. Indeed, phase I trials with 8-Cl-cAMP have been performed ([ 19 , 20 ] and reviewed in [ 21 ]) and phase II trials are in progress [ 22 ]. However, the mechanisms by which this agent acts are incompletely understood, and several investigators have proposed that an 8-Cl-cAMP metabolite, 8-chloro-adenosine (8-Cl-Ado) is the active anti-proliferative compound [ 16 , 23 ]. Indeed, 8-Cl-Ado has been shown to inhibit growth in a variety of cell types [ 24 - 28 ]. Previously, we demonstrated that 8-Cl-Ado arrests the growth of primary mouse epidermal keratinocytes without triggering differentiation [ 29 ]. Thus, 8-Cl-Ado functions in an analogous fashion to transforming growth factor-β (TGF-β), which also triggers growth arrest, but not differentiation,, of keratinocytes (reviewed in [ 30 ]). In contrast with a polypeptide such as TGF-β, 8-Cl-Ado, as a small molecule rather than a protein, could potentially be taken orally or applied topically to skin. Thus, 8-Cl-Ado may represent a novel therapy for treatment of skin disorders, such as psoriasis, actinic keratoses and basal and squamous cell carcinomas, characterized by hyperproliferation of keratinocytes. One potential problem, however, is that TGF-β also inhibits the expression of differentiation markers elicited by other differentiating agents [ 31 ]. Since another characteristic typical of hyperproliferative skin diseases such as psoriasis is impaired differentiation [ 32 ], a therapy that inhibits both proliferation and differentiation would be less than ideal. To determine whether 8-Cl-Ado, as a potent keratinocyte growth arrestor, could potentially be used to treat hyperproliferative skin diseases in combination with a current treatment, we investigated the effect of 8-Cl-Ado on 1,25(OH) 2 D 3 -induced inhibition of keratinocyte proliferation and stimulation of keratinocyte differentiation. We found that low concentrations of 8-Cl-Ado acted additively with 1,25(OH) 2 D 3 to inhibit DNA synthesis, without affecting the ability of 1,25(OH) 2 D 3 to enhance keratin 1 expression, a marker of early differentiation, or transglutaminase activity, a marker of late differentiation. Thus, our results suggest that a combination therapy with 1,25(OH) 2 D 3 and 8-Cl-Ado could potentially be an effective treatment for hyperproliferative skin disorders including psoriasis, actinic keratosis and non-melanoma skin cancers. Results and discussion To determine if 8-Cl-Ado could function with the growth inhibiting agent 1,25(OH) 2 D 3 to enhance its antiproliferative effect, we incubated primary epidermal keratinocytes for 24 hours with various concentrations of 8-Cl-Ado in the presence and absence of low concentrations of 1,25(OH) 2 D 3 prior to assessing effects on de novo DNA synthesis as measured by [ 3 H]thymidine incorporation into DNA. As shown in Figure 1A , 8-Cl-Ado inhibited [ 3 H]thymidine incorporation at concentrations of 5–25 μM with an estimated half-maximal inhibitory concentration (IC 50 ) of 5 μM. This value agrees well with our previously determined IC 50 of 7.5 μM [ 29 ]. In agreement with previous reports [ 33 , 34 ], 1,25(OH) 2 D 3 also inhibited DNA synthesis at concentrations of 1 to 100 nM with an estimated IC 50 of approximately 4 nM (Figure 1B ). As shown in Figure 2 , when the two agents were combined, their effect on DNA synthesis appeared to be additive, as evidenced by the comparable slopes of the [ 3 H]thymidine incorporation curves at the three different concentrations of 0 (a portion of which is replotted from Figure 1 ), 1 and 10 nM 1,25(OH) 2 D 3 . The combination of 1 or 5 μM 8-Cl-Ado with 10 nM 1,25(OH) 2 D 3 yielded a greater inhibition than 8-Cl-Ado alone, and conversely, the combined effect of 5 and 10 μM 8-Cl-Ado with 1 nM 1,25(OH) 2 D 3 was significantly larger than 1 nM 1,25(OH) 2 D 3 alone. Importantly, the combination of 10 nM 1,25(OH) 2 D 3 with 10 μM 8-Cl-Ado produced an inhibition of [ 3 H]thymidine incorporation that was significantly greater than that elicited by either agent alone. Indeed, the inhibition elicited by 10 μM 8-Cl-Ado and 10 nM 1,25(OH) 2 D 3 was comparable to the inhibition produced by 100 nM 1,25(OH) 2 D 3 alone (compare Figures 1B and 2 ). Thus, our results suggest that not only does 8-Cl-Ado not prevent the growth inhibitory action of 1,25(OH) 2 D 3 , but, in fact, the two agents seem to act in an additive fashion to more effectively inhibit keratinocyte proliferation. Figure 1 8-Cl-Ado and 1,25(OH) 2 D 3 Inhibit Keratinocyte Proliferation. Near-confluent primary mouse epidermal keratinocytes were treated with the indicated concentrations of (A) 8-Cl-Ado or (B) 1,25(OH) 2 D 3 for 24 hours, and [ 3 H]thymidine incorporation was determined as indicated in Materials and Methods. Data represent the mean ± SEM of five experimentsperformed in triplicate; p < 0.05, p < 0.01 versus the control value. Figure 2 8-Cl-Ado and 1,25(OH) 2 D 3 Act Additively to Inhibit Keratinocyte Proliferation. Near-confluent primary mouse epidermal keratinocytes were treated with the indicated concentrations of 8-Cl-Ado in the presence of no (closed circles), 1 nM (open squares) or 10 nM (open triangles) 1,25(OH) 2 D 3 for 24 hours, and [ 3 H]thymidine incorporation was determined as indicated in Materials and Methods. Data represent the mean ± SEM of five experiments performed in triplicate; p < 0.05, *p < 0.01 versus the control value, †p < 0.01 versus the corresponding concentration of 1,25(OH) 2 D 3 alone, §p < 0.01 versus the corresponding concentration of 8-Cl-Ado alone. TGF-β, another agent that, like 8-Cl-Ado, induces growth arrest but not differentiation of keratinocytes ([ 31 ] and reviewed in [ 30 ]), can inhibit the ability of differentiating agents to elicit keratinocyte differentiation [ 31 ]. However, for an agent to have therapeutic potential as a treatment for hyperproliferative skin disorders, such an inhibition of differentiation would be counterproductive to its efficacy as a medication. To determine if 8-Cl-Ado also inhibited keratinocyte differentiation, we investigated whether 8-Cl-Ado inhibited the ability of 1,25(OH) 2 D 3 to induce the late differentiation marker, transglutaminase activity. For this experiment we chose the concentrations of 8-Cl-Ado (10 μM) and 1,25(OH) 2 D 3 (10 nM) shown in Figure 2 to produce a greater growth inhibition than either agent alone. As illustrated in Figure 3 , 10 μM 8-Cl-Ado alone had little or no effect on transglutaminase activity, as reported previously [ 29 ]. On the other hand, 10 nM 1,25(OH) 2 D 3 significantly elevated transglutaminase activity by approximately 75%. The combination of 8-Cl-Ado and 1,25(OH) 2 D 3 was not significantly different from 1,25(OH) 2 D 3 alone, with a significant approximate 60% increase relative to the control value. Thus, our results indicate that 8-Cl-Ado did not prevent the differentiative effect of 1,25(OH) 2 D 3 , suggesting that these two agents might be combined to treat keratinocyte hyperproliferative disorders, such as psoriasis. Figure 3 8-Cl-Ado Has No Effect on 1,25(OH) 2 D 3 -Stimulated Transglutaminase Activity. Near-confluent primary mouse epidermal keratinocytes were treated with and without 10 μM 8-Cl-Ado in the presence and absence of 10 nM 1,25(OH) 2 D 3 for 24 hours, and transglutaminase activity was determined as indicated in Materials and Methods. Data represent the mean ± SEM of four experiments performed in triplicate; p < 0.01 versus the control value. Transglutaminase activity is a marker of late keratinocyte differentiation. We also examined the effect of 8-Cl-Ado on a marker of early keratinocyte differentiation, namely keratin 1 protein expression, using an even higher concentration of 8-Cl-Ado (25 μM). Western analysis demonstrated that 1,25(OH) 2 D 3 induced an approximate 45% increase in keratin 1 protein levels with the combination of 1,25(OH) 2 D 3 and 8-Cl-Ado producing a comparable 46% increase (Figure 4 ). Thus, early differentiation in response to 1,25(OH) 2 D 3 also was not affected by 8-Cl-Ado. Interestingly, however, in contrast to previous results [ 29 ], in these experiments 8-Cl-Ado alone elicited a small but significant increase in keratin 1 protein expression (32%). The reason for this disparity is unclear but may result from differences in the lot of anti-keratin 1 antibody used in the western analysis and/or the increased sensitivity of the method used for detecting and quantifying immunoreactive protein in this work. Figure 4 8-Cl-Ado Has No Effect on the 1,25(OH) 2 D 3 -Induced Increase in Keratin 1 Protein Levels. Near-confluent keratinocytes were incubated for 24 hours with and without 25 μM 8-Cl-Ado in the presence and absence of 20 nM 1,25(OH) 2 D 3 and were then processed for western analysis. (A) A representative immunoblot is shown. (B) Keratin 1 levels were quantified, corrected for background and normalized for loading, as described in Materials and Methods. Data represent the mean ± SEM of four experiments performed in duplicate; *p < 0.05 versus the control value. Most current treatments for psoriasis suffer from one or more disadvantages including lack of efficacy, contraindications due to deleterious side effects and/or aesthetic deficiencies ([ 35 ] and reviewed in [ 36 ]). Indeed, monotherapies tend to be less efficacious than combination therapies with two or more agents used concurrently, sequentially or in a rotational fashion (reviewed in [ 36 ]). Treatment with 1,25(OH) 2 D 3 and its analogs has proven successful, although the possibility of toxicity as the result of 1,25(OH) 2 D 3 's ability to affect calcium metabolism has led to the search for topically effective analogs with little or no effect on serum calcium levels (reviewed in [ 32 ]). If the amount of 1,25(OH) 2 D 3 (or its analog) required for treatment could be reduced, this decrease in dosage would presumably minimize systemic effects on calcium, which is the primary dose-limiting factor in the use of 1,25(OH) 2 D 3 analogs in the treatment of psoriasis [ 32 ]. Thus, our results indicating that 8-Cl-Ado enhances the growth inhibitory effect of 1,25(OH) 2 D 3 , a known keratinocyte differentiating agent and possible treatment for psoriasis [ 32 ], suggests the potential for combination therapy. Moreover, the fact that 8-Cl-Ado does not interfere with the promotion of differentiation by 1,25(OH) 2 D 3 further supports the possible combined use of these two agents for treatment of hyperproliferative skin disorders. Several lines of evidence suggest that 8-Cl-Ado is not simply acting through cyototoxicity to inhibit keratinocyte growth. First, we have previously shown that 8-Cl-Ado growth arrests keratinocytes in the G 0 /G 1 phase of the cell cycle with no increase in the sub-G 0 /G 1 (apoptotic) population of cells [ 29 ]. Second, we also showed that the effect of 8-Cl-Ado to inhibit proliferation is reversible in that washout of the compound returned DNA synthesis essentially to basal (untreated) levels [ 29 ]. Finally, in this report we demonstrate that 8-Cl-Ado did not inhibit the 1,25(OH) 2 D 3 -stimulated increase in transglutaminase activity (Figure 3 ) or keratin 1 protein expression (Figure 4 ). Together, these results indicate that 8-Cl-Ado is acting in a specific manner to decrease keratinocyte proliferation. Nevertheless, the mechanism by which 8-Cl-Ado exerts its growth inhibitory effects in keratinocytes is not clear. Our previous results indicate that 8-Cl-Ado must enter the cells to trigger growth arrest, since inhibiting uptake with an adenosine transporter, NBTI, prevented the arrest in the G 0 /G 1 phase of the cell cycle [ 29 ]. We also reported in a prior publication that 8-Cl-Ado induced the expression of the cyclin-dependent kinase inhibitor, p21 [ 29 ], which is known to contribute to growth arrest in keratinocytes and other cell types ([ 37 ] and reviewed in [ 30 ]). However, other investigators have reported 8-Cl-Ado-mediated inhibitory effects on RNA synthesis and the levels of cellular ATP [ 16 ]. Clearly, further research is necessary to define the pathways used by 8-Cl-Ado to regulate keratinocyte proliferation. Conclusions In summary, our data show that 8-Cl-Ado functions with the keratinocyte-differentiating agent 1,25(OH) 2 D 3 to inhibit keratinocyte proliferation without altering the ability of 1,25(OH) 2 D 3 to induce differentiation. Thus, our results support the possibility of using 8-Cl-Ado alone or in combination with differentiating agents such as 1,25(OH) 2 D 3 or its analogs to treat hyperproliferative keratinocyte disorders including psoriasis. Methods Materials Tissue culture reagents were obtained from standard suppliers as indicated in a previous publication [ 29 ]. 1,25(OH) 2 D 3 was a generous gift of Dr. Maurice Pechet (Research Institute for Medicine and Chemistry, Cambridge, MA). 8-Cl-Ado was obtained from Biolog (La Jolla, CA). [ 3 H]Thymidine and [ 3 H]putrescine were purchased from Dupont/NEN (Boston, MA). Dimethylated casein was obtained from Sigma (St. Louis, MO). All other reagents were from standard suppliers. Keratinocyte culture Primary cultures of mouse epidermal keratinocytes were prepared from neonatal ICR CD-1 mice and cultivated in a 25 μM calcium-containing serum-free keratinocyte medium as in [ 29 ]. Measurement of DNA synthesis For measurement of [ 3 H]thymidine incorporation into DNA, as in [ 29 ], near-confluent cultures were refed with SFKM containing various concentrations of 8-Cl-Ado with or without different concentrations of 1,25(OH) 2 D 3 . After 24 hours, cells were labeled with 1 μCi/ml [ 3 H]thymidine for an additional hour in the continued presence of 8-Cl-Ado and/or 1,25(OH) 2 D 3 . Cultures were washed twice with phosphate-buffered saline without calcium or magnesium (PBS - ) and macromolecules were precipitated using ice-cold 5% trichloroacetic acid (TCA). After additional washing with 5% TCA and distilled water, cells were solubilized in 0.3 M NaOH, and the amount of [ 3 H]thymidine incorporated into DNA was determined by liquid scintillation counting. Measurement of transglutaminase activity Transglutaminase activity was assessed essentially as described in [ 33 ]. Briefly, near-confluent keratinocytes were incubated for 24 hours with the indicated agents in SFKM. The cells were scraped into homogenization buffer (0.1 M Tris-acetate, pH 7.8, 2 μg/ml aprotinin, 2 μM leupeptin, 1 μM pepstatin A, 0.2 mM EDTA and 0.2 mM PMSF), collected by centrifugation and subjected to one freeze-thaw cycle prior to disruption by sonication. Aliquots of the homogenate were removed for determination of protein content and transglutaminase activity. Transglutaminase activity was measured as the [ 3 H]putrescine radioactivity incorporated into casein after an overnight incubation at 37°C. Casein was precipitated with TCA, collected onto glass fiber filters and counted by liquid scintillation spectrometry. The cellular protein content of the samples was determined using the Bio-Rad DC protein assay system (Bio-Rad, Hercules, CA), with BSA as standard, and transglutaminase activity was expressed as cpm/μg protein. Western analysis of keratin 1 protein levels Keratinocytes were treated and solubilized in sample buffer (31.2 mM Tris, pH 6.8, 1% SDS, 12.5% glycerol). Equal sample volumes were separated by SDS polyacrylamide gel electrophoresis on an 8% gel and transferred to Immobilon PVDF membrane (Millipore, Billerica, MA). Membranes were blocked with Odyssey blocking buffer (Licor Biosciences, Lincoln, NE), probed with a rabbit polyclonal anti-keratin 1 antibody (Covance, Princeton, NJ) and a mouse monoclonal anti-actin antibody (Sigma, St. Loius, MO). Immunoreactive proteins were visualized with IRDye800-coupled donkey anti-rabbit IgG (Rockland Immunochemicals, Gilbertsville, PA) or IR Alexa Fluor 680-coupled goat anti-mouse IgG (Molecular Probes, Eugene, OR) on a Licor Odyssey Infrared Imaging System. Keratin-1 protein levels were corrected for background and normalized using background-corrected actin levels. Statistical analysis Significance of differences was determined with the computer program InStat (Graphpad Software, San Diego, CA) using ANOVA with a Student-Newman-Keuls post-hoc test. Abbreviations 1,25(OH) 2 D 3 , 1,25-dihydroxyvitamin D 3 ; 8-Cl-Ado, 8-chloro-adenosine; 8-Cl-cAMP, 8-chloro-cyclic-adenosine monophosphate; IC 50 , half-maximal inhibitory concentration; TGFβ, transforming growth factor-beta Authors' contributions WBB conceived of the study, planned the experiments, analyzed the data and drafted the manuscript; XZ and SJ planned, conducted and analyzed the keratin 1 expression experiments.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC509244.xml
538270	Comparative study of matrix metalloproteinase expression between African American and Caucasian Women	To date there are 26 human matrix metalloproteinases (MMPs) which are classified according to their substrate specificity and structural similarities. The four major subgroups of MMPs are gelatinases, interstitial collagenases, stromelysins, and membrane-type matrix metalloproteinases (MT-MMPs). This study investigates the expression of 26 MMPs, which have been shown to play a role in cancer metastasis. Breast tissues and cell lines derived from African American patients and Caucasian patients were assayed to demonstrate alterations in the transcription of genes primarily responsible for degrading the extracellular matrix (ECM). The expression levels of the extracellular matrix and adhesion molecules were analyzed using the gene array technology. Steady state levels of mRNAs were validated by RT-PCR analysis. Total RNA was isolated from tissue and cell lines and used in the RT-PCR assays. From this data, differential expression of MMPs between 6 breast cancer cell lines and 2 non-cancer breast cell lines was demonstrated. We have performed an in vitro comparison of MMP expression and established differences in 12 MMPs (3, 7, 8, 9, 11–15, 23B, 26, and 28) expression between African American and Caucasian breast cell lines. Thus, evidence indicates that altered expression of MMPs may play a role in the aggressive phenotype seen in African American women.	Introduction In 2003, it was estimated that approximately 1.3 million Americans would be diagnosed with invasive cancer. Of this group, racial/ethnic minorities account for a disproportionate number of these cancers [ 1 , 2 ]. Invasive breast cancer usually begins in either the lobules or the ducts of the breast. These tumors then metastasize via the breast associated and thoracic lymphatic tissue [ 3 ]. The incidence of breast cancer in Caucasian women (112 out of 100,000) is higher than in African American women (AA) (95 out of 100,000) after the age of 40, however, the mortality suffered by (AA)(37 out of 100,000) is higher than Caucasian women (CAU) (31 out of 100,000) at every age [ 4 ]. Thus a greater percentage of AA women die from breast cancer and resulting metastasis. In 2004, an estimated 215,990 new cases of invasive breast cancer is expected to occur among women in the United States [ 5 ]. Breast cancer is the most common cancer among AA women; however, the rate of newly diagnosed cases is about 13% lower than CAU women [ 6 ]. There is accumulating evidence that AA women have a higher frequency of more aggressive tumor types, which have been shown to lead to higher mortality rates. Studies show that compared Caucasian women (CAU), African American women (AA), regardless of age had proportionally more Grade III tumors and fewer Grade I and II tumors for all stages combined and for each individual stage group [ 7 ]. The grade of cancer has been shown to be a prognostic factor with higher-grade tumors being associated with reduced survival [ 7 ]. The most common cause of death in breast cancer patients is metastasis of breast cancer cells to bones, lungs, liver and brain and the progressive growth of the cancer at these sites [ 7 , 8 ]. Therefore, controlling breast cancer metastasis represents an effective method of preventing or slowing disease progression. The extracellular matrix (ECM) is a complex structural entity surrounding and supporting organs and tissues of the body. The ECM plays a key role in cell-cell signaling, wound repair, cell adhesion and tissue function. Recent studies suggest that cell adhesion proteins located on breast cancer cells interact with the ECM [ 7 ]. This interaction induces increased production by the breast cancer cells of proteins that degrade the ECM. This degradation enables the tumor cells to invade the surrounding tissue and ultimately enter the circulatory system. Once they are in circulation, tumor cells travel to other organ sites where they progressively grow [ 7 ]. The matrix metalloproteinases (MMPs) are a family of structurally and functionally related endoproteinases that are involved in the degradation of the ECM. Currently, there are 26 identified human matrix metalloproteinases, which are classified according to their substrate specificity and structural similarities [ 8 ]. Abnormal expression of these proteins contributes to various pathological processes including rheumatoid arthritis and tumor growth, invasion and metastasis. The four main subgroups of MMPs are the interstitial collagenases, which catalyze degradation of fibrillar forms of collagen, the gelatinases which degrade gelatin and collagen that are abundant in basement membranes, the stromelysins, which degrade various substrates including proteoglycans, laminin and collagen I, II, and III and the membrane-type MMPs which have been shown to catalyze activation of progelatinase A, to degrade a variety of ECM substrates and to function as a fibrinolytic enzyme in the absence of plasmin [ 9 ]. MMP expression has bee shown to be elevated during development, pregnancy, and involution and has been shown to be related to tumor cell invasiveness [ 10 ]. This study investigates the expression levels of the 26 identified MMPs, which have been shown to play a role in the metastatic process using breast tissues and cell lines derived from AA and CAU women. Materials and Methods Cell Culture and Tissue RNA All cell lines were purchased from American Type Culture Collection (Rockville, MD, USA). Cells were propagated in the recommended media and given new media every 2 to 3 days until 90% confluent (see table 1 ). Human Breast Tissue RNA was purchased from Ambion (Austin, TX). RNA Extractions RNA was extracted from the cell line using the RNAqueous (Ambion, Austin, TX). Cells were collected by low speed centrifugation and lysed by adding 200 μl of Lysis/Binding Solution. An equal volume of 64% ethanol was added to the lysate. The lysate/ethanol mixture was transferred to the RNAqueous Filter Cartridge and centrifuged for 1 minute at 13,400 rpm. The flow through was discarded and 700 μl of Wash Solution 1 was added to the RNAqueous Filter Cartridge and centrifuged for 1 minute. The column was washed twice with 500 μl of Wash Solution 2/3 and eluted with 110 μl Elution Solution. Isolated RNA was quantitated using the UltraSpec 2000 (Pharmacia Biotech). All RNA samples were stored at -70°C in RNA elution solution until further use. Gene array The Extracellular Matrix and Adhesion Molecule gene arrays were obtained from SuperArray (Frederick, MD). The array membranes were pre-hybridized with GEA hybridization solution and denatured salmon sperm DNA at 60°C for two hours. For each RNA sample, a labeling mix consisting of 4 μl 5X GEA labeling buffer, 2 μl biotin-16-dUTP, 1 μl RNase inhibitor, 1 μl reverse transcriptase, and 2 μl RNase-free water was prepared and an aliquot of 3 μg of RNA was added to each respective thin-walled PCR tube. The cDNA labels were created using a cycle of 3 minutes at 70°C, 2 minutes at 42°C, and an additional 90 minutes at 42°C. Two microliters of stop buffer was added and the mix denatured at 94°C for 5 minutes. The labeled cDNA was added to the membrane and allowed to hybridize overnight. The membranes were washed with 2X SSC/0.1% SDS and 0.1X SSC/0.5% SDS, blocked with blocking solution, and the probes were detected using AP-Strepavidin, specific buffers, CDP-Star and subsequent exposure to X-ray film for 30 seconds to 5 minutes. The autoradiograms were analyzed using ScanAlyzer and GEArray Analyzer (SuperArray, Frederick, MD). Reverse Transcriptase Polymerase Chain Reaction (RT-PCR) The RT-PCR reactions were performed in a P/E GeneAmp 9700 thermocycler (Perkin-Elmer Co., Norwalk CT), using the Access RT-PCR system (Promega, Madison, WI). The reaction mixes were prepared by combining 27. 5 μl of nuclease free water, 10 μl of AMV, 1 μl Tfl 5X reaction buffer, 2 μl dNTP mix, 50 pM of upstream primer, 50 pM of downstream primer in 1.5 μl volume each, 3 μl 25 mM MgSO 4 , 1.0 μl AMV reverse transcriptase, Tfl DNA polymerase and 1 μg of total RNA in a 0.5 ml thin walled Eppendorf tube on ice. The reaction mixes were then vortexed for 5 seconds and centrifuged. The PCR cycling profile was as follows: 48°C for 1 minute for reverse transcription of the RNA into cDNA, 94°C for 4 minutes to deactivate the reverse transcriptase, and 30 cycling sequences of denaturing at 94°C for 45 seconds, annealing at 55°C for 30 seconds, and extension at 72°C for 1 minute with a final extension at 72°C for 10 minutes. An aliquot of 20 μl of each RT-PCR reaction were run on 1.2% agarose gels, stained with ethidium bromide, photographed and subjected to densitometic measurements using the Chemi-Imager Tm 4000 (Alpha Innotech, Corporation, San Leandro, CA). Results Gene Array Analysis Gene arrays were utilized to explore and compare the expression levels of extracellular matrix adhesion molecules in AA and CAU breast cancer cells. The array revealed elevated expression in 36% of the genes in AA samples when compared to their CAU counterparts. Of those elevated genes, 31% were from the cell membrane adhesion molecules group, 17% from the extracellular matrix proteins functional gene group, 37% from the proteases category, and 14% were protease inhibitors. Initial results of the gene array indicated a significant elevation of the proteases (data not shown). To further evaluate this, we proceeded with direct analysis of all known MMPs. MMP RT-PCR Analysis Comparison of the individual relative densities between AA and CAU women revealed elevated expression in 12 of the 26 MMPs (3, 7, 8, 9, 11, 12, 13, 14, 15, 23B, 26, and 28) (Figure 1 and Table 2 ). Elevated expression of MMP-3, 7, 8, 9, 11, 12, 13, 14, 23B, 26 and 28 in AA breast cancer cells was observed when the overall averages of the expression levels of all AA and CAU women cell lines were compared (Table 3 ). Figure 1 RT-PCR expression of MMPs in African American and Caucasian breast cell lines and tissue. Table 2 Matrix Metalloproteinase Expression Assessment by RT-PCR African American Normal Caucasian Cell Lines 2315 2320 2329 A1N4 10A MCF7 Hs578t 2336 MMPs 1 6 ± 2.7 12 ± .58 11 ± 1.15 5 5.7 6 ± 2.7 15 ± 2 11 ± 2.1 2 45.3 ± 4.2 32 ± .58 13.3 ± .58 6.7 11 11.7 ± 7.8 59 ± 1 33 ± 11 3 53 ± 11.5 47.3 ± 11 64 ± 7 9 14.3 26.5 18.3 40 7 73.3 ± 2.3 40.67 ± 2.1 17.6 ± 2.5 5.3 45.3 9 ± 1.7 38 ± 14.53 36.7 ± 4 8 14 ± 2.5 17 ± 2.5 16 ± 2.1 5 6 10 ± 1 15 ± 1.2 12 ± 1 9 66 ± 3 43 ± 5.2 18 ± 5.5 42.3 74.3 7.3 ± 1.5 28 ± 2.7 20.3 ± 5.13 10 7.3 ± .58 26 ± 1.7 18.7 ± 5.5 5.3 5 9.3 ± 1.15 66.3 ± 4 11.3 ± 1.5 11 11 ± 2 15.7 ± 2.9 15 ± 5 5 4 10 ± 4 13 ± 3.79 14 ± 0.58 12 10 ± 4 18 ± 8.9 16 ± 3.6 4 7 5 ± 1.15 5 ± 1 7 ± 2.1 13 10 ± 2 21.7 ± 4.2 7.7 ± 1.2 4 73.67 8.3 ± 2.1 18.7 ± 3.5 6.7 ± 1.15 14 12 ± 3 13.7 ± 1.5 16 ± 5 7 10 13 ± 8.5 11 ± 1.5 14 ± 3.1 15 70 ± 2.7 30 ± 0.58 57.7 ± 6.3 18 7.67 46.3 ± 4.5 81.3 ± 7.5 26.3 ± 6.8 16 8.7 ± 4 7.7 ± 1 15.7 ± 4 7.67 7.67 16 ± 2.9 16 ± 1 8 ± 2.7 17 N/D N/D N/D N/D N/D N/D N/D N/D 19v1 4 ± 1 7.3 ± .58 5 ± 1 5.33 4.33 5.3 ± 1.5 4.7 ± .58 11 ± 3.5 19v3 16 ± 6 13.3 ± 2.9 21.3 ± 8.7 15.3 10 18.3 ± 6.1 24 ± 4.2 25 ± 5 19v6 9.7 ± 4.6 9 ± .58 8 ± 5.2 7 6 12 ± 2.1 11 ± 2.1 9 ± 2.7 19v9 10 ± 1.0 10 ± 3.5 11 ± 2.0 7 7 10 ± 5 10 ± 2.0 10 ± 2.3 20 10.3 ± 3.2 45 ± .58 12.3 ± 2.1 11 9.3 23.7 ± 12.4 37.7 ± 6.0 11 ± 1 23A 5.7 ± 1.5 12 ± 2.0 8.7 ± 1.5 7 7 10 ± 2.8 5.7 ± .58 6.3 ± .58 23B 7 ± 1.1 32 ± 10.6 19.3 ± 1.5 3 7.3 10 ± 1.4 9.7 ± 1.5 12.7 ± 3.2 24 33 ± 1.2 19.7 ± 1.2 36 ± 1 18.67 30.2 33.7 ± 1 35 ± .58 37 ± 1.5 25 8 ± 1 12 ± 5.2 9 ± 1.0 5 6 13 ± 1.5 12.7 ± 1.0 7 ± 2.0 26 11 ± 1.0 11 ± 1.7 9 ± 2.5 11 8 10 ± 2 7 ± .58 9 ± 1.0 27 19 ± 7.2 19 ± 6.0 18 ± 8.0 6.7 11.67 15 ± 2.0 20 ± 13.2 19 ± 1.0 28 9 ± 5.6 14 ± 5.7 7 ± 2.0 15 18 7 ± .58 10 ± .58 9 ± 6.0 RT-PCR expression of MMPs in AA and CAU cells. Elevated expression in AA vs. CAU denoted in bold. Mean ± SD N/D: not detected Table 3 Averaged Relative Density of MMP Expression For AAW CAU and Normal Cell Lines AAW CAU Normal MMPs 1 3.2 3.4 5.35 2 9.6 11.5 11.3 3 18.3 8.4 11.6 7 14.63 9.3 25.3 8 5.37 4.18 5.5 9 14.1 6.2 58.3 10 5.8 9.7 5.1 11 4.62 4.2 4.5 12 4.9 2 5.5 13 4.4 3.7 38.8 14 4.63 4.3 8.5 15 17.5 17.1 12.8 16 3.6 4.5 7.67 17 N/D N/D N/D 19v1 1.8 2.3 4.83 19v3 5.6 7.5 12.65 19v6 3.0 3.6 6.5 19v9 3.4 3.4 7 20 7.5 8.0 10.1 23A 2.9 2.4 7 23B 6.5 3.7 5.1 24 9.9 11.7 24.4 25 3.2 3.7 5.5 26 3.5 2.9 9.5 27 6.1 6 9.1 28 3.3 2.96 16.5 RT-PCR expression of MMPs in AA, CAU cancer cell lines and Normal cell lines.. Elevated expression in AA -vs- CAU denoted in bold. N/D: not detected Discussion Little is known as to why the incidence of breast cancer is lower yet mortality is higher in African American women. Many studies speculate that this is only a socio-economical problem [ 11 ]. However this investigation provides another possibility that may reveal molecular mechanisms that contribute to the increased mortality of AA women with breast cancer. The major threat to patients with breast cancer is tumor invasion and metastasis [ 12 ]. Tumor invasion is a complex process that requires interaction between the invasive cells and the ECM [ 13 ]. This process involves a cascade of events including angiogenesis, local invasion, and intravasation. One of these critical steps involves the proteolytic degradation of the ECM and basement membrane. This is partially done by the matrix metalloproteinases. One aspect related to cancer progression has been considered in numerous studies is the association of MMP expression with tumor grade and aggressiveness [ 14 ]. The GEArray Q Series Human Extracellular Matrix and Adhesion Molecules Gene Array were used to determine the expression profiles of various types of matrix and adhesion molecules. The array was divided into four components: cell adhesion molecules, extracellular matrix proteins, proteases and protease inhibitors. From analysis of the gene array, altered expression was observed in many of the proteases. These findings led to further study of the matrix metalloproteinases (data not shown). Gene Array analysis of AA and CAU breast cancer cells indicates that there is altered expression of the genes in the Extracellular matrix and adhesion molecules, particularly the proteases. This group included 17 MMPs of which ten displayed elevated expression in AA women. RT-PCR was performed to confirm the results of the gene array. We observed elevated expression of 12 MMPs in AA cell lines when compared to their CAU counterparts. These include one gelatinase (MMP-9), two interstitial collagenases (MMP-8, and 13), 3 stromelysins (MMP-3, 7, 11), two MT-MMPs (MMP-14 and -15) and 4 uncategorized MMPs (MMP-12, 23B, 26, and 28). There was no MMP-17 expression detected in any of the cell lines (Figure 1 ). Studies have shown that normal mammary gland expression of MMPs is low except during times of development, pregnancy, and involution [ 10 , 15 , 16 ]. However, during pathologic states such as breast cancer, increased levels of MMPs have been reported in breast tumor cells as well as in the surrounding non-cancerous breast tissue [ 17 ]. Our results suggest that there is altered expression of MMPs in cell lines derived from AA and CAU women. It also demonstrates that there is greater expression of MMPs in AA women than in CAU women. This investigation indicates that altered expression of MMPs may play a role in the aggressive phenotype seen in AA women. This evidence suggests that the elevated expression levels of 12 MMPs may be a contributing factor in the higher mortality rates of AA breast cancer patients. This study is significant because it may reveal biomarkers of metastasis in AA women. To date, this is the first study to extremely investigate MMP expression in cell lines derived from African American patients. Abbreviations used MMP-Matrix metalloproteinases, MT-MMP-Membrane-type matrix metalloproteinases, AA-African American women, CAU-Caucasian, RT-PCR-Reverse transcriptase Polymerase Chain Reaction, ECM-Extracellular matrix Author's Contributions JAM performed the microarrays and RT-PCR experiments, was involved in tissue culture and prepared the first draft of the manuscript. HFY was responsible for primer design, and performed data analysis and densitometric readings of the gene arrays and RT-PCR. KL maintained all cells and tissues and assisted in the editing of this manuscript. MJ provided cell lines, training in microarray performance and editing. AAD conceived the study and participated in its design, coordination and funding, as well as preparation of the manuscript. All authors read and approved the final manuscript. Table 1 Cell Lines and Tissue Samples Human Breast Tissue Normal breast tissue (derived from CAU) MCF-10A Mammary gland, fibrocystic disease (CAU) A1N4 Mammary epithelial, chemically transformed (CAU) CAUCASIAN (CAU) HS578T Mammary gland; breast; carcinoma MCF-7 Mammary gland; breast; epithelial; metastatic site: pleural effusion adenocarcinoma CRL-2336 Mammary gland epithelial, primary ductal carcinoma AFRICAN AMERICAN (AA) CRL-2315 Breast, primary ductal carcinoma CRL-2329 Carcinoma, ductal, primary; breast; mammary gland CRL-2320 Carcinoma, ductal, breast; mammary gland; from metastatic site: lymph node	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC538270.xml
529427	Reconsidering Early HIV Treatment and Supervised Treatment Interruptions	Another study casts doubt on the value of early treatment and treatment interruptions. What are the implications of this study for our understanding of HIV pathogenesis, treatment, and vaccine development?	The devastating effects of HIV infection worldwide are reason enough for AIDS researchers to grasp at thin rays of hope. But seldom has a single anecdotal case stimulated as much hope as the 1999 report of an acutely infected patient who appeared to control HIV replication after two short treatment interruptions [ 1 ]. This report generated the hypothesis that early antiretroviral treatment (during or very soon after symptomatic seroconversion) allows the incompletely damaged immune system to recover and respond appropriately to virus antigens during treatment interruptions. This, in turn, according to the hypothesis, leads to control of viral replication by a healed and appropriately stimulated immune response to the patient's HIV infection. Consistent with this hypothesis was the prior finding that early antiretroviral therapy led to induction of HIV-specific proliferative responses similar to those that had been observed in patients with long-term, non-progressing HIV [ 2 ]. This led Rosenberg and colleagues to ask whether HIV-specific proliferative responses were a necessary and sufficient cause of long-term non-progression or just an immunologic consequence of controlled virus replication. Their report of virologic control in patients who interrupted therapy after early treatment raised hope that if HIV infection was treated early enough, the immune system could be repaired sufficiently to allow for long-term immunologic control of HIV replication [ 3 ]. Unfortunately, that's where the good news ends. Enthusiasm Fades A series of discoveries from clinical trials began to chip away at the enthusiasm for both early treatment of HIV infection and supervised treatment interruptions (STIs) as a way to boost the immune response. Several small trials of STIs in chronically infected patients were carried out [ 4 ], buoyed by the reasonable desire of patients for respite from the unpleasant side effects of the drugs. These trials gave disappointing results, up to and including the emergence of antiretroviral drug resistance in patients randomized to receive STIs. HIV-specific immune responses did increase off therapy, but so did viral loads. The so-called immune boosting probably reflected an immune response to greater viral antigen load but did not represent constructive immune enhancement. Larger trials clearly showed that STIs were of little if any benefit in chronic infection and that when therapy was stopped, viral loads invariably returned to pre-treatment levels [ 5 ]. Other studies indicated that HIV-specific CD4+ T cells were being preferentially infected, often massively, during treatment interruptions [ 6 ], and that proliferative responses were more likely to be a consequence—rather than a cause—of decreased HIV replication [ 7 ]. Despite multiple attempts, early reports of an inverse correlation between simple HIV-specific T cell responses and virologic control were not confirmed [ 8 ]. Where complex T cell functions did show such a correlation, the data indicated that viral replication was adversely affecting the character of the T cell immune response to HIV, and not the other way around [ 9 ]. Thus, no evidence of “immune boosting” during STIs and subsequent viral control in the absence of antiretroviral drugs was ever established. Finally, one of the acutely treated patients within Rosenberg's cohort became superinfected with a second strain of HIV despite excellent control of viral replication and significant recognition of the superinfecting strain by the pre-existing T cell response [ 10 ]. STIs offered patients hope of respite from taking complex regimens, but trials have been disappointing (Photo: J Troha) New Findings Now comes a study in this month's PLoS Medicine that found that in 14 patients who were treated early and who had controlled viral loads for at least 90 days, the virologic control was only transient [ 11 ]. While one could look at this as a glass half full—these patients achieved a reasonable period of time off antiretroviral therapy—closer scrutiny of the data limits this view. There was a disconnect between the low viral loads and an unexpectedly high rate of CD4+ T cell decline in several patients. While the small number of patients and the single-arm nature of the study preclude definitive comparisons, it is possible that the early treatment and STIs did not result in a delay in CD4+ T cell decline (and, therefore, initiation of antiretroviral therapy) beyond what would have occurred had the patients received no early treatment. Implications of the Study This study raises important questions in our understanding of HIV pathogenesis, treatment, and vaccine development. First, why is it that early antiretroviral treatment, even if it does lead to better control of viral replication, does not protect against CD4+ T cell depletion? It is possible that by the time patients present with acute retroviral syndrome their CD4+ T cell reserves (in gut and lymphoid tissues) have been severely depleted, despite the fairly normal CD4+ profile of their peripheral blood. Thus, even low-level viral replication is then sufficient to deplete the remaining central and peripheral reserves [ 12 ]. Second, how do these findings affect treatment guidelines during acute infection? None of the current treatment guidelines in either resource-rich or resource-poor settings recommend early antiretroviral therapy. In the light of these new data [ 11 ], there does not appear to be a rationale for early antiretroviral therapy in the absence of a clinical trial to assess other interventions in concert with early therapy. The use of therapeutic vaccination is an obvious intervention that still needs to be tested, despite limited efficacy results in treated chronic infection. As such, practice guidelines should continue to caution against early treatment unless associated with a randomized clinical trial. Finally, is this good or bad news for HIV vaccine development? Since most current vaccine strategies are based upon the hypothesis that induction of T cell immunity will lead to control of viral replication, it is difficult to be optimistic when a strong and broad immune response is unable to prevent disease progression. However, one must recall that phenotypic and functional assessments of HIV-specific T cell responses, even in antiretroviral-treated patients, show that these responses clearly differ from responses against viruses that are normally cleared or controlled by the immune system [ 9 ]. Therefore, the T cell responses in the patients treated for acute HIV infection in Kaufmann et al.'s study were induced upon a dramatically altered immune background. It remains to be determined how much this adversely affects the HIV-specific immune response, and whether an immune response generated by vaccination before any HIV replication (a prophylactic vaccine) might be better able to control virus replication. Far be it for us to stop grasping at rays of hope.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC529427.xml
538258	Immunological parameters in girls with Turner syndrome	Disturbances in the immune system has been described in Turner syndrome, with an association to low levels of IgG and IgM and decreased levels of T- and B-lymphocytes. Also different autoimmune diseases have been connected to Turner syndrome (45, X), thyroiditis being the most common. Besides the typical features of Turner syndrome (short stature, failure to enter puberty spontaneously and infertility due to ovarian insufficiency) ear problems are common (recurrent otitis media and progressive sensorineural hearing disorder). Levels of IgG, IgA, IgM, IgD and the four IgG subclasses as well as T- and B-lymphocyte subpopulations were investigated in 15 girls with Turners syndrome to examine whether an immunodeficiency may be the cause of their high incidence of otitis media. No major immunological deficiency was found that could explain the increased incidence of otitis media in the young Turner girls.	Introduction Recurrent otitis media is often a problem in children with Turner syndrome (TS) [ 1 , 2 ]. More than 60% of the Turner girls (60–80%) aged 4–15 years suffer from repeated attacks of acute otitis media, as compared to 5% of children (aged 0–6 years) in the normal population [ 3 , 4 ]. These problems among the Turner girls are more extensive and last longer (up in their teens) than in an non Turner population. Frequent insertions of myringeal tubes are often necessary and in order to try to prevent chronic ear problems regular and frequent controls are necessary. However, sequelae like chronic otitis media are frequently seen, even if controls have been meticulous. A sensorineural hearing loss is also common among these patients, with a typical dip in the mid frequencies, declining over time. This sensorineural dip has been identified already in 6-year-old Turner girls [ 3 ]. Later in life (~35 years) a progressive high frequency hearing loss is added to the dip, leading to more prominent hearing problems and hearing aids often become necessary [ 2 , 5 , 6 ]. The cause of the associated ear and hearing problems is not known but the ear problems later in life could be influenced by the loss of estrogen. TS is caused by the presence of only one normally functioning X-chromosome. The other sex chromosome can be missing (45, X) or abnormal and mosaicism is often present. Occurring in one of every 2000 female births, TS is one of our most common sex chromosome abnormalities [ 7 ]. TS is characterized by short stature, no spontaneous puberty and infertility due to ovarian dysgenesis with no estrogen production [ 8 ]. Mental retardation is not connected to the syndrome. Since the early 80's, treatment is given with growth hormone from birth and estrogen therapy to induce puberty. Immunological disturbances have previously been described in TS, with an association to reduced levels of serum IgG and IgM, increased IgA and decreased levels of circulating T- and B-lymphocytes. However, the results have not been conclusive [ 9 - 12 ]. In the normal population children with IgG 2 deficiency commonly develop recurrent acute otitis media. It is believed that these infections are secondary to impaired antibody response, rather than Eustachian tube dysfunction [ 13 ]. As immunological derangements seem to be common in TS, an immunological deficiency could be a potential cause to parts of the ear problems. The aim of this study was to investigate immunoglobulin and lymphocyte subpopulations in girls with Turners syndrome to examine whether an immunodeficiency may be the cause of their high incidence of otitis media. Immunotherapy would then be a possible treatment. Materials and methods Subjects Blood samples from patients with the diagnosis TS, genetically confirmed, were investigated according to the Swedish ethical record no 88–265. Analyses regarding immunoglobulin- and lymhpocyte subpopulations were performed in 15 girls, aged 5–17 years (median age 11 years), randomly selected from all girls in this age group with TS attending the Karolinska Hospital, Stockholm (total 29 patients). Of these 53% (n = 8) had suffered from repeated attacks of otitis media. All TS girls had been treated with growth hormones and their karyotypes were: 45, X (n = 8); 45, X/46, XX (n = 4); 45, X/46, X, i(Xq) (n = 2); and 45, X/46, X, r(X) (n = 1) (r = ring chromosome). A medical history was attained, focusing on autoimmune diseases, previous and current ear diseases and other infectious diseases, ear operations, and hearing problems. Lymphocyte subpopulations Leukocyte counts (10 9 /L) were analysed in a Coulter MicroDiff II (Beckman-Coulter). The differential leukocyte (lymphocytes, monocytes and granulocytes) counts and percentages were obtained by 2-color FACS-analysis with CD14/CD45 markers. The number and percentage of lymphocyte subpopulations were obtained by standardized 2- or 3-color FACS-analysis on Epics XL or Elite flowcytometer (Beckman-Coulter) using commercial reagents. CD19 + was marker for B-cells and CD3 + for T-cells, CD3 + CD4 + for helper T-cells, CD3 + CD8 + for cytotoxic T-cells, CD56 + CD3 - for NK-cells and HLA-DR + for activated T-cell subsets. The ratio of CD4 + /CD8 + was also calculated. The monoclonal antibody clones used were: UCHT1 (CD3 + ), SFCI12T4D11/T4 (CD4 + ), SFCI21Thy2D3/T8 (CD8 + ), 116/Mo2 (CD14 + ), 89B/B4 (CD19 + ), KC56 (CD45 + ), NKH1 (CD56 + ) and 9-49/I3 (HLA-DR + ), all from Cytostat, Beckman-Coulter. All FACS-analyses were performed at the routine laboratory, Department of Clinical Immunology, Karolinska Hospital and the results were compared to age-related in-house and published reference ranges (5 to 95 percentiles) [ 14 ] except for CD56 + CD3 - for which an adult reference was used (10–90 percentile). Complement and antibodies Hemolytic complement (classical and alternative pathways), IgA antibodies to gliadin and endomysium, IgG antibodies to pneumococcal polysaccharide and tetanus toxoid antigen, the serum concentrations (g/L) of circulating IgA, IgG, IgM, IgD, IgG 1 , IgG 2 , IgG 3 and IgG 4 as well as the Gm(23)-allotyping of IgG 2 were analysed by standard methods and compared to age related reference ranges used at the routine laboratory, Department of Clinical Immunology, Karolinska Hospital, Stockholm. Statistical analysis Medians of continuous parameters were compared between groups by Mann-Whitney U-test and correlations were performed by Spearman rank analysis. A two-tailed p < 0.05 was considered significant. Results Lymphocyte subpopulations The leukocyte counts as well as the absolute counts and percentages of lymphocytes, monocytes, and granulocytes were within normal limits for all 15 Turner girls. Likewise most girls had normal counts and percentages of lymphocyte subpopulations as compared to the 5 to 95% percentiles age-related reference ranges (Fig. 1a and 1b ) including activated CD4 + and CD8 + T-cells (HLA-DR + ). However, the CD4 + /CD8 + ratio was in the lower range (girls aged ≥10), with one girl having a very low ratio (0.6). Figure 1 1a and b Percentages (Fig 1a) and absolute counts (Fig 1b) of lymphocyte subpopulations in 15 girls with Turner's syndrome divided into two age groups. Group A aged <10 years (n = 4) and group B aged ≥10 years (n = 11). Girls with recurrent otitis media are illustrated with open symbols (n = 8) and those who are otitis free with filled symbols (n = 7). The horizontal lines indicate medians and the shaded boxes the 5 to 95 percentiles of age-related reference ranges except for CD56 + CD3 - cells for which the 10 to 90 percentiles reference range of adults was used. Complement and Immunoglobulin levels Hemolytic complement (classical and alternative pathway) was within normal limits for all 15 Turner girls. The serum concentrations of IgG, IgA, IgM, IgD and the four IgG subclasses were for most Turner girls within the age-related 95% confidence intervals (Fig. 2 ). The exceptions were one girl with elevated IgM (2.3 g/L), five with elevated IgD (0.1–0.23 g/L), two with elevated IgG 1 (10.2 and 10.8 g/L), one with low IgG 2 (0.4 g/L) and two girls with low IgG 4 (<0.01 g/L). Figure 2 Immunoglobulin levels in 15 Turner girls. The shaded boxes indicate the 95% confidence interval for the 5–20 years age group. Girls with recurrent otitis media are illustrated with open symbols (n = 8) and those who are otitis free with filled symbols (n = 7). The frequency of homozygous G2m(23)-negative Turner girls was 33% (5/15). Antibodies Normal levels of IgG antibodies to tetanus toxoid and polysaccharide antigen were detected among most Turner girls, except for two respectively one, having too low levels. Slightly elevated IgA antibodies to gliadin were observed in 3 (20%) girls, whereas no IgA antibodies to endomysium could be detected in any of the 15 girls. Age When comparing girls aged <10 years (n = 4) and ≥10 years (n = 11) the following parameters were found to be influenced by age with decreased values among the older girls: total counts of leukocytes (p = 0.0093), lymphocytes (p < 0.05), monocytes (p = 0.0093), granulocytes (p = 0.015), CD19 + (p = 0.0053) and CD4 + HLA-DR + (p = 0.035), as well as the percentage of CD19 + (p = 0.023). Also IgG 2 increased with age (p = 0.05). These findings are in line with the reference literature for the normal population [ 14 ]. Recurrent Otitis Media The girls with TS were divided into two groups according to their history of recurrent otitis media. As age influenced some of the parameters we only considered girls ≥10 years old (n = 11). Significant increases in absolute counts of lymphocytes (p = 0.004), CD3 + T-cells (p = 0.0087), CD4 + T-cells (p = 0.012) and CD4 + HLA-DR + (p = 0.05) as well as in the percentage of CD3 + T-cells (p = 0.05) in otitis prone (n = 5) compared to otitis free (n = 6) Turner girls was shown. No such differences were noticed for any immunoglobulin levels, antibody titers, CD4 + /CD8 + -ratio or CD8 + , CD19 + , CD56 + CD3 - lymphocyte subpopulations. Karyotype Any apparent influence, of the different karyotypes, on any of the parameters studied was not observed within the group. Discussion In this study no major derangement in the immune status was found among the girls with TS. Normal levels of most lymphocyte- and immunoglobulin subpopulations were registered. The few outliers noted must be considered as a normal individual variation. However, as described in an earlier study of Turner girls, the present study confirmed a CD4 + /CD8 + ratio in the lower range [ 12 ], supposedly as a consequence of a slightly increased CD8 + population. Although, the patients were few, we noticed some differences between the otitis prone and otitis free Turner girls. The elevated counts of lymphocytes, CD3 + , CD4 + cells and CD4 + HLA-DR + cells seen among the otitis prone girls, probably reflects a secondary effect of an activated immune system involving T-helper cells, rather than any immune deficient state. Moreover, the levels of IgG antibodies to pneumococcal polysaccharide antigen, which are important in the defense of bacteria, were normal. A homozygous lack of the IgG2m(23) allotype was seen in 33% of the girls, which is the same frequency as in the normal population [ 15 ]. A negative IgG2m(23) allotype have been correlated to an impaired immune response to haemophilus influenzae vaccination with subnormal levels of IgG 2 . In the study group a negative IgG2m(23) allotype was not correlated to a positive history of recurrent otitis media, neither could the different karyotypes be associated to the levels of immunoglobulin- or lymphocyte subpopulations. Perhaps the cause of the repeated attacks of otitis media in Turners syndrome is not to be found in the periphery, but rather more locally. Even if earlier computed tomography scans of the temporal bone have not shown any abnormalities [ 2 ], the Eustachian tube may be dysfunctional and/or the cell system might be underdeveloped. Recently new aspects on the growth of the temporal bone have been proposed, with a hypothesis that the loss of X-chromosome material leads to a prolonged cell cycle and otic growth disturbances during fetal life [ 16 ]. The SHOX-gene located on the p-arm of the X-chromosome has been found to code for growth and could potentially also code for growth of the skull base and temporal bone where the middle ear is located. [ 17 ]. As the girls investigated were 5–17 years old, transient hypogammaglobulinemia in the first years is still possible. However, the girls suffered otitis media up in their teens. Our findings of normal immunoglobulin- and lymphocyte subpopulations are not entirely in concordance with some earlier studies, where a reduction of circulating IgM and IgG as well as T- and B-lymphocytes has been observed [ 9 , 10 ]. However, in these studies the values were not dramatically decreased, but rather within the lower range of the normal reference values. On the other hand, some other studies have not shown low T- and B-lymphocyte counts [ 11 ] or low concentrations of immunoglobulins [ 12 ], agreeing with the present study. In the normal population there is a difference between IgG and IgM levels in women and men with decreased values in men [ 12 ], but this difference cannot be found in newborns or children. Earlier there have been suggestions that the difference is caused by the amount of X chromosome material, as men with 47, XXY have higher values than men with normal karyotype (46, XY) and women with 47, XXX have even higher values than normal women (46, XX) [ 18 ]. There have also been suggestions that the sex hormones influence the immune system and that the lack of estrogens might influence the immune response negatively [ 11 ]. As most of the girls studied were prepubertal, the influence from sex hormones should not be as important. In some earlier studies the age span has been wider and the size of the study groups relatively small. There have also been discussions that the regular treatment with growth hormones may influence the immune system. However, in a previous study no major effects on the immunoglobulin levels or lymphocyte subpopulations could be demonstrated in Turner girls treated with growth hormones [ 12 ]. In conclusion, we did not find any major immunological deficiency in immunoglobulins or lymphocyte subpopulations that could explain the increased incidence of otitis media observed in girls with TS. Therefore, treatment with immunotherapy is not an option in this patient group. Further studies are warranted to elucidate local pathology, both from an immunological and anatomical point of view. Authors' contributions AES participated in the design of the study, performed the statistical analysis and drafted the manuscript. LS participated in the design of the study and collected the blood samples. CGMM performed the statistical analysis. MH participated in the design and coordination of the study and collected the blood samples. All authors read and approved the final manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC538258.xml
514567	Thermal modeling of lesion growth with radiofrequency ablation devices	Background Temperature is a frequently used parameter to describe the predicted size of lesions computed by computational models. In many cases, however, temperature correlates poorly with lesion size. Although many studies have been conducted to characterize the relationship between time-temperature exposure of tissue heating to cell damage, to date these relationships have not been employed in a finite element model. Methods We present an axisymmetric two-dimensional finite element model that calculates cell damage in tissues and compare lesion sizes using common tissue damage and iso-temperature contour definitions. The model accounts for both temperature-dependent changes in the electrical conductivity of tissue as well as tissue damage-dependent changes in local tissue perfusion. The data is validated using excised porcine liver tissues. Results The data demonstrate the size of thermal lesions is grossly overestimated when calculated using traditional temperature isocontours of 42°C and 47°C. The computational model results predicted lesion dimensions that were within 5% of the experimental measurements. Conclusion When modeling radiofrequency ablation problems, temperature isotherms may not be representative of actual tissue damage patterns.	Introduction The mitigation of primary and metastatic tumors by radiofrequency ablation is a developing research area. The goal of ablation is to necrose treatment volumes by raising the temperature of targeted tissues. Ablation probes are inserted percutaneously, laparoscopically, or during surgery into cancerous tumors. Once positioned, high frequency alternating current (450–550 kHz) is delivered through an uninsulated electrode into the surrounding tissues to a dispersive ground pad that is applied to the patient. The electromagnetic energy is converted to heat by resistive heating. While the usage of radiofrequency ablation devices is well established, efforts to optimize treatment strategies are ongoing. An important consideration in optimizing ablation is determining what treatment volumes are necessary and acceptable. In liver ablation, for example, treatment volumes generally extend a centimeter beyond the dimensions of a tumor [ 1 - 3 ]. Since the liver possesses regenerative characteristics, it is more critical to insure that necrosis is achieved in 100% of the cancerous cell volume than to minimize damage to healthy tissues. In contrast, a centimeter margin in cardiac ablation is generally unacceptable since many vital substructures are in close proximity. The growth of ablation lesions remains a central issue in the development of radiofrequency ablation devices. Knowing the expected shape of lesions is essential for treatment planning and procedure optimization. To date, many approaches have been attempted to characterize lesion size. The results have varied widely. Ablation lesions generated in vitro and in vivo in animal models show wide variations, since many of the key parameters (i.e. tissue perfusion) cannot be controlled [ 4 , 5 ]. In addition, the boundaries of lesions in animal models are often "fuzzy" and are subject to interpretation. Computational modeling is a valuable tool in the optimization process, since it allows the systematic examination of the various parameters affecting the outcome of ablation. However, most computational models fail to capture essential physiologic phenomena. Many computational studies have been reported in the literature to predict the growth of lesion size during ablation [ 6 - 19 ]. However, the majority of these models do not directly calculate lesion size. Surrogate endpoints, such as temperature [ 20 - 22 ] and thermal dosing [ 23 ] are calculated and are interpreted as being equivalent to lesion size. In many cases, these surrogate endpoints do not correlate well with clinical outcome and vary considerably. The microwave hyperthermia literature, for example, cites 42 degrees Celsius as the point at which thermal damage occurs to tissues [ 24 , 25 ]. In the cardiac ablation literature, 47 degrees Celsius is generally accepted as the onset of tissue damage [ 23 , 24 , 26 , 27 ]. Neither of these values can be derived directly from gross histological measurements of lesion size, since the tissue pathology does not provide a record of temperature. Many computational studies justify these surrogate endpoints by showing a high correlation between temperature isotherms and lesion size. However, temperature isotherms and lesion size have never actually been shown to be equivalent. Several investigators have demonstrated that tissue damage is a function of both temperature and time [ 28 - 30 ]. As tissue temperature is increased, the amount of time necessary to achieve a threshold of damage decreases. Tissue damage can be characterized using the Arrhenius equation which relates temperature and exposure time using a first order kinetics relationship. Data from experimental studies, where tissues are exposed to uniform temperatures for controlled time intervals, are fit to the Arrhenius equation to determine the frequency factor A (s -1 ), and the activation energy Δ E (J mol -1 ). Arrhenius parameters have been determined in skin [ 31 - 35 ], artery [ 36 , 37 ], blood [ 38 - 40 ], pancreas [ 41 , 42 ], heart [ 43 ], cornea [ 44 - 46 ], muscle [ 47 ], prostate [ 48 ], ovary [ 49 ], kidney [ 50 - 52 ], and liver [ 30 , 52 , 53 ]. For a specified exposure temperature and time, the fit parameters A and Δ E determine the amount of cell damage incurred for a specific tissue type. In combination with computational modeling techniques, it is then possible to calculate the distribution of cell damage surrounding ablation probes. In this study, we compare the temperature distribution and tissue necrosis patterns for a hepatic ablation probe at body temperatures. At each time step, the specific absorption rate (SAR), temperature, and the tissue damage are calculated. The level of tissue perfusion is varied for the models to determine the maximum variation in lesion size resulting from a typical ablation. These data are validated experimentally using an ablation probe in liver tissue. Methods Radiofrequency ablation probes operate between 460–550 kHz. At these frequencies, the wavelength of the electromagnetic energy is several orders of magnitude larger than the size of the ablation electrodes. Thus, the primary mode of energy transfer is through electrical conduction and can be modeled as a coupled quasistatic electrical conduction and heat conduction problem. The electric field is solved by using Laplace's equation, ∇·[ σ ( T )∇ V ] = 0 (Eq.1) where ∇ is the gradient operator, σ (T) is the temperature-dependent conductivity (Siemens/meter), and V is the electric potential (Volts). Temperature is solved by using a modified Pennes bioheat equation [ 54 ], where ρ is the density, 1060 kg/m 3 [ 55 ]; C is the heat capacity of tissues, 3600 J/kg-K [ 55 ]; k is the heat conduction coefficient, 0.502 W/K-m [ 55 ]; ρ b is the density of blood, 1000 kg/m 3 [ 9 ]; C b is the heat capacity of blood, 4180 J/kg-K [ 9 ];α is a tissue state coefficient; ω is the blood perfusion coefficient, 6.4 × 10 -3 sec -1 [ 9 ];T amb is the ambient body temperature, 37°C; and Q m is the metabolic heat source term. For all cases, we assumed that the metabolic heat source was insignificant. The tissue state coefficient (α) ranges from 0–1 depending on the local level of tissue damage At each time step, the cumulative damage integral is computed using the well established Arrhenius equation where Ω(t) is the degree of tissue injury, c(t) is the concentration of living cells, R is the universal gas constant, A is a "frequency" factor for the kinetic expression (s -1 ), and Δ E is the activation energy for the irreversible damage reaction (J-mol -1 ) [ 50 ]. The kinetic parameters account for morphologic changes in tissue relating to the thermal degradation of proteins [ 56 ]. The parameters A and Δ E are dependent on the type of tissue and have been characterized for liver tissues by Jacques et. al. (A = 7.39 × 10 39 s -1 and ΔE = 2.577 × 10 5 J-mol -1 ) [ 52 ]. In the context of finite element modeling of tissue damage, a damage integral of Ω = 1, corresponds to a 63% percent probability of cell death at a specific point. A damage integral of Ω = 4.6, corresponds to 99% percent probability of cell death at a point in the model. The significance of Ω = 1 has been reported as the point at which tissue coagulation first occurs [ 36 ]. Once tissue coagulation occurs, tissue perfusion ceases. This corresponds to a tissue state coefficient of α = 0. Intermediate levels of the tissue state coefficient are calculated as α = 1/exp(Ω). Figure 1 shows a diagram of a typical needle ablation electrode used in clinical practice for hepatic tumor ablation. The probe is 6.0 cm long with a diameter of 0.15 cm. The distal 2.0 cm of the probe is uninsulated and the proximal 4.0 cm of the probe is covered with a thin electrically insulating material. Figure 2 shows a three-dimensional representation of the axisymmetric two-dimensional geometry of the model. The active portion of the probe is situated in the center of a cylindrical model that is 6.0 cm in radius and 12.0 cm in height. Electrical and thermal properties of liver are used in the model to simulate a fully-embedded insertion of the needle electrode. The electrical properties of tissue are assumed to be temperature dependent and solved according to Chang [ 57 ], where the electrical conductivity appears as Figure 1 Ablation probe geometry diagram of a single needle ablation electrode that is used for hepatic tumor ablation. Therapeutic treatment is achieved by applying a source voltage to the conducting tip. A conducting pad applied to the patient skin serves as an electrical ground return. Figure 2 Model geometry three dimension representation of the axisymmetric two-dimensional finite element model. All external surfaces of the cylindrical model serve as the electrical ground and are at body temperatures (37°C). The entire ablation probe is assumed to be thermally insulating. σ(T, N ) = σ(25, N ) {1.000-1.962 × 10 -2 Δ + 8.08 × 10 -5 Δ 2 - N Δ [3.020 × 10 -5 + 3.922 × 10 -5 Δ + N (1.721 × 10 -5 Δ - 6.584 × 10 -6 Δ)]} (Eq.4) where σ (25, N ) = N [10.394-2.3776 N + 0.68258 N 2 - 9.13538 N 3 + 1.0086 × 10 -2 N 4 ] ;N is the normality of an electrically equivalent sodium chloride solution, N = 0.0111; and Ä = 25-T, which produces an equivalent electrical conductivity of liver tissues at 37°C (approximately 0.134 S/m). The thermal properties of liver used in the model were acquired from Tungjitkusolmun et al. [ 9 ] and Duck [ 55 ]. A source voltage (V o ) is applied to the conducting tip of the ablation probe. The outer surface of the model serves as an electrical ground return (V = 0). An electrically insulating boundary condition is applied to the non-conducting portions of the probe such that n ·(σ∇ V) = 0; where n is the unit vector normal to the surface, σ is the electrical conductivity, and V is the voltage at the insulating surface. A thermal boundary condition of T = T amb is applied to the outer surfaces of the model to simulate ambient temperature. Since the thermal mass of the probe is small compared to the surrounding tissue, we assumed that heat conduction into the probe itself was minimal. Thus, all other surfaces of the ablation probe are considered to have a thermally insulating boundary condition such that n ·(k ∇ T) = 0. A hybrid finite element model was developed using Femlab (Comsol, Burlington MA, USA) and Matlab (Mathworks, Natick MA, USA) to calculate temperature and tissue damage. While conventional finite element models effectively solve field solutions using a nonuniform geometrical mesh, tissue exposure calculations are integrated at each point in the model over the course of ablation and are more easily calculated using uniform rectilinear grids. As shown in Figure 3 , Femlab is used to solve the coupled electromagnetic and heat conduction equations simultaneously at each timestep. This is done to insure that the temperature-dependent electrical conductivity is updated with each iterative calculation of temperature for a given timestep. The converged temperature is mapped from the finite element mesh into a rectilinear grid, which is passed into the Matlab environment. The amount of tissue damage occurring at each timestep is calculated using the Arrhenius equation and tracked at each point in the model. Once the level of damage exceeds 63% cell damage, it is assumed that tissue coagulation has occurred, causing a cessation in tissue perfusion. The 63% cell damage point is historically used because it corresponds to the earliest onset of visible tissue coagulation. A rectilinear grid containing the perfusion characteristics at each point in the model is mapped back into the finite element mesh and used in subsequent Femlab calculations. The rectilinear grid of temperature is also used to calculate the change in the electrical conductivity which is an explicit function of temperature. This data is also mapped back into the finite element mesh and is used to change the electromagnetic sourcing characteristics. Augmented matrices are used to insure that calculations made on the geometric borders of the Femlab model are interpolated correctly. Figure 3 Computational technique diagram of data flow used in a hybrid finite element model implemented in Femlab/Matlab to calculate temperature and tissue damage. The electric field and temperature are solved simultaneously in Femlab (blue blocks). The data structure is changed from finite element meshing to rectilinear gridding so that the resulting temperature can be used to calculate tissue exposure and electrical conductivity change in Matlab (orange blocks). A tissue damage level of 63% corresponds to the onset of tissue necrosis and is associated with a cessation in local blood flow. The Matlab results are then imported into Femlab as inputs for calculation at the next time step. Given the axial symmetry of the problem, we used a 2D-axisymmetric mesh consisting of 13,641 nodes and 26,880 elements. The Femlab 'Fldaspk' ordinary differential equation solver was used to achieve convergence. This is a robust variant of the traditional ODE15s stiff differential equation solver used in solving finite element problems in Matlab. Ablations were simulated at source voltages of 0, 2.5, 5, 7.5, 10, 12.5, 15, 17.5, 20, 22.5, 25, 27.5, and 30 volts. For each of the source voltages, we varied the initial level of tissue perfusion at 0%, 20%, 40%, 60%, 80%, and 100% normal tissue perfusion (6.4 × 10 -3 cubic meters of blood/ cubic meter of tissue/ second) [ 9 ]. Ambient tissue temperature was assumed to be 37°C. The model simulates a 15 minute ablation and updates tissue parameters at 2 second timestep intervals. Once the 15 minute ablation has ended, the model continues to solve solutions for 15 minutes post-ablation. For each simulation, the electric field (E), the current density (J), the temperature (T), and the tissue damage (D) were calculated. All calculations were implemented on a Dual 3.02 GHz Xeon processor workstation with 4 GByte RAM. Each simulation takes approximately 3 hours to run. Experimental Validation To validate the computational model, experimental measurements were made in 6 freshly excised porcine liver sections. A single needle ablation probe with a 2 cm uninsulated tip was inserted 3 cm into each liver tissue. Since commercial RF ablation generators operate using either constant temperature or constant power feedback algorithms, an experimental constant voltage RF generator (500 kHz) was used [ 5 ]. Tissue samples were allowed to equilibrate to room temperature (approximately 22°C) prior to the start of ablation. Two samples were ablated at 20 volts for 15 minutes. After allowing the tissue to cool for an additional 15 minutes, the probes were removed and the tissue was bisected to expose the lesion. The tissues were immediately placed in a 1% 2,3,5-triphenyltetrazolium chloride (red) solution for 20 minutes to stain for tissues containing active dehydrogenase, an indicator of cell viability [ 58 , 59 ]. This stains healthy tissue brick red, leaving the ablated region a pale grey color. The maximum width and depth of the macroscopically pale ablated regions were measured. This procedure was repeated for two samples at 25 volts and for the last two samples at 30 volts. Computational model calculations were made at 20, 25, and 30 volts following the same experimental protocol. Ambient temperature for these calculations was 22°C instead of the 37°C temperature used in the main simulations. The calculated lesion sizes were directly compared with the measurements in tissue. Results Table 1 shows the maximum temperatures attained in tissue for the computational models for a range of voltages (2.5–30 Volts) and tissue perfusion rates (0–100% normal tissue perfusion.). The table shows a nonlinear relationship between the source voltage and the maximum temperature that results from the use of a temperature-dependent electrical conductivity. The maximum variation in the temperature data for a given source voltage did not exceed 17%. The data show that the rate of temperature increase accelerates as a function of the source voltage. As the level of tissue perfusion increases, tissue temperature decreases. Table 1 Maximum Temperature (Degrees Celsius) 1 Values represent the maximum temperature attained in tissue for the computational models. Source Voltage (Volts) 0% 2 Perfusion (0.0 × 10 -3 m b 3 /m t 3 /s) 20% 2 Perfusion (1.3 × 10 -3 m b 3 /m t 3 /s) 40% 2 Perfusion (2.6 × 10 -3 m b 3 /m t 3 /s) 60% 2 Perfusion (3.8 × 10 -3 m b 3 /m t 3 /s) 80% 2 Perfusion (5.1 × 10 -3 m b 3 /m t 3 /s) 100% 2 Perfusion (6.4 × 10 -3 m b 3 /m t 3 /s) 2.5 37.3 37.3 37.2 37.2 37.2 37.2 5.0 38.6 38.5 38.4 38.3 38.2 38.2 7.5 41.0 40.6 40.3 40.1 40.0 39.9 10.0 44.3 43.6 43.1 42.8 42.5 42.3 12.5 48.8 47.7 46.9 46.3 45.8 45.4 15.0 54.6 52.9 51.6 50.7 50.0 49.4 17.5 62.0 59.5 57.7 56.2 55.2 54.3 20.0 71.1 67.9 65.5 63.4 61.9 60.7 22.5 82.4 78.6 75.4 73.0 70.7 69.0 25.0 96.1 91.8 88.0 84.8 82.1 79.7 27.5 112.7 3 107.8 3 103.4 3 99.5 96.1 93.4 30.0 132.5 3 126.9 3 121.8 3 117.4 3 113.4 3 109.9 3 1 – The maximum temperature for the case of 0% perfusion was located along the center of the conducting electrode. In all other cases, the maximum temperature occurred at the tip of the probe; 2 – The units for tissue perfusion are cubic meters of blood (m b 3 ) per cubic meter of tissue (m t 3 ) per second; 3 – These temperatures do not account for energy loses associated with tissue desiccation or gas formation. Table 2 shows the maximum electrical conductivity in the tissue after heating for 15 minutes at a variety of source voltages. All tissues initially have an electrical conductivity of 0.144 S/m at 37°C. The data show that tissue electrical conductivity is primarily a function of the source voltage, changing 320% over the course of a 15 minute ablation using a 30 volt source. With normal tissue perfusion (6.4 × 10 -3 m b 3 / m t 3 /s), the electrical conductivity changes as much as 260% using a 30 volt source. The electrical conductivity is indirectly a function of tissue perfusion since tissue perfusion is zero in the necrosed treatment volume. Tissue perfusion lowers the tissue temperature outside the treatment volume which helps to conduct heat away from temperatures within the ablated area.. Table 2 Maximum Electrical Conductivity (Siemens/meter) 1 Values represent the maximum electrical conductivity attained in tissue for the computational models. Source Voltage (Volts) 0% 2 Perfusion (0.0 × 10 -3 m b 3 /m t 3 /s) 20% 2 Perfusion (1.3 × 10 -3 m b 3 /m t 3 /s) 40% 2 Perfusion (2.6 × 10 -3 m b 3 /m t 3 /s) 60% 2 Perfusion (3.8 × 10 -3 m b 3 /m t 3 /s) 80% 2 Perfusion (5.1 × 10 -3 m b 3 /m t 3 /s) 100% 2 Perfusion (6.4 × 10 -3 m b 3 /m t 3 /s) 2.5 0.144 0.144 0.144 0.144 0.143 0.143 5.0 0.147 0.147 0.146 0.146 0.146 0.146 7.5 0.153 0.152 0.151 0.151 0.150 0.150 10.0 0.162 0.160 0.159 0.158 0.157 0.156 12.5 0.173 0.170 0.168 0.167 0.165 0.164 15.0 0.189 0.185 0.181 0.179 0.177 0.175 17.5 0.210 0.203 0.198 0.194 0.191 0.189 20.0 0.238 0.228 0.221 0.215 0.210 0.207 22.5 0.274 0.262 0.251 0.244 0.237 0.232 25.0 0.321 0.306 0.293 0.282 0.273 0.265 27.5 0.383 0.364 0.348 0.334 0.321 0.312 30.0 0.463 0.440 0.419 0.401 0.385 0.372 1 – The maximum electrical conductivity for the case of 0% perfusion was located along the center of the conducting electrode. In all other cases, the maximum temperature occurred at the tip of the probe; 2 – The units for tissue perfusion are cubic meters of blood (m b 3 ) per cubic meter of tissue (m t 3 ) per second. Table 3 shows the maximum SAR computed for a range of voltages and tissue perfusion rates. The SAR is defined as SAR = σ /ρ*\|E\| 2 , where σ is the electrical conductivity, ρ is the tissue density, and \|E\| is the magnitude of the electric field. The data shows that the SAR is highest with increasing source voltage with no tissue perfusion. Initially, this seems counterintuitive as one would expect a higher maximum SAR for perfused flows, where a greater amount of power is needed to compensate for the convective heat loss. This observation can be explained by the large changes in the electrical conductivity (Table 2 ). Since higher temperatures are achieved for cases with no tissue perfusion, the change in the electrical conductivity is highest with no tissue perfusion. Since, at a given point, the density and the magnitude of the electric field are essentially constant (<0.02% change), the SAR will vary as a function of the electrical conductivity only. Table 3 Maximum Specific Absorption Rate (Watts/kg) 1 Values represent the maximum specific absorption rate (SAR) attained in tissue for the computational models. Source Voltage (Volts) 0% 2 Perfusion (0.0 × 10 -3 m b 3 /m t 3 /s) 20% 2 Perfusion (1.3 × 10 -3 m b 3 /m t 3 /s) 40% 2 Perfusion (2.6 × 10 -3 m b 3 /m t 3 /s) 60% 2 Perfusion (3.8 × 10 -3 m b 3 /m t 3 /s) 80% 2 Perfusion (5.1 × 10 -3 m b 3 /m t 3 /s) 100% 2 Perfusion (6.4 × 10 -3 m b 3 /m t 3 /s) 2.5 645.6 645.3 645.1 644.9 644.8 644.8 5.0 2608 2603 2600 2598 2596 2595 7.5 5966 5940 5924 5912 5903 5896 10.0 10850 10770 10720 10680 10650 10630 12.5 17450 17260 17120 17030 16950 16900 15.0 26030 25620 25330 25120 24970 24860 17.5 36940 36140 35600 35190 34900 34680 20.0 50590 49180 48230 47480 47000 46620 22.5 67470 65210 63520 62510 61610 60990 25.0 88170 84890 82360 80440 78940 77910 27.5 113300 108900 105300 102400 100100 98390 30.0 143400 137700 133000 129000 125700 123000 1 – The maximum current density was always located at the tip of the ablation probe. Therefore, all values listed above are comparable with each other; 2 – The units for tissue perfusion are cubic meters of blood (m b 3 ) per cubic meter of tissue (m t 3 ) per second. Figure 4 shows the tissue temperature and the cell death penetration into tissue for a 15 minute ablation using a 30 volt constant voltage source for perfusion rates ranging from no perfusion to 100% normal tissue perfusion. The data show that cell death decreases more rapidly than tissue temperature. At the center of the active electrode, temperatures decrease as a function of the inverse of the radius squared (1/r 2 ), whereas cell damage exhibits an S-shaped curve. Figure 4 shows that 63% tissue damage is roughly correlated with the 60°C isotherm for liver tissues. Conventional temperature isotherms for tissue damage for hyperthermia (42°C) and radiofrequency ablation (47°C) substantially overestimate the size of the lesions. Figure 4 Tissue temperature and cell death penetration for a 15 minute ablation using a 30 volt constant voltage source. Simulation results for a 15 minute ablation using a 30 volt constant voltage source measured from the center of the active electrode. The graph shows temperature (solid) and cell death (dotted) penetration into liver tissue for a range of tissue perfusion rates. The units for tissue perfusion are cubic meters of blood (m b 3 ) per cubic meter of tissue (m t 3 ) per second (s). Figure 5 shows a plot of tissue temperature and cell damage calculated at a distance of 4 millimeters from the center of the active electrode for a 15 minute ablation using a 30 volt constant voltage source. Temperature decrease and cell damage that occurs after the ablation is monitored for an additional 15 minutes. The data show that near the electrode, tissue damage will reach 100% well within the first few minutes of energy application. For cases of no tissue perfusion, 100% tissue damage occurs after 5 minutes at a distance of 4 millimeters. For cases with normal tissue perfusion, 100% tissue damage occurs approximately 8 minutes into the ablation. At a distance of 10 millimeters from the center of the active electrode under the same conditions (Figure 6 ), the data show that tissue damage will not always reach 100%. For the case of no perfusion, 100% cell damage is reached a minute after the termination of radiofrequency energy. In cases with varying levels of tissue perfusion, cell damage is significantly reduced and, in some cases, insignificant. Although the overall temperatures are lower at 10 millimeters than at 4 millimeters, temperatures near 60°C are reached but do not result in complete tissue damage because the length of time in which the tissue is exposed is not sufficient. Figure 5 Tissue temperature and cell death at a distance of 4 millimeters from the center of the active electrode using a 30 volt constant voltage source. Ablation simulation results attained 4 millimeters from the center of the active electrode for a 15 minute ablation using a 30 volt constant voltage source. The graph shows temperature (solid) and cell death (dotted) penetration into liver tissue for a range of tissue perfusion rates. The units for tissue perfusion are cubic meters of blood (m b 3 ) per cubic meter of tissue (m t 3 ) per second (s). Figure 6 Tissue temperature and cell death at a distance of 10 millimeters from the center of the active electrode using a 30 volt constant voltage source. Ablation simulation results attained 10 millimeters from the center of the active electrode for a 15 minute ablation using a 30 volt constant voltage source. The graph shows temperature (solid) and cell death (dotted) penetration into liver tissue for a range of tissue perfusion rates. The units for tissue perfusion are cubic meters of blood (m b 3 ) per cubic meter of tissue (m t 3 ) per second (s). Figure 7 and 8 show comparisons of temperature distribution and lesion size development with no tissue perfusion (Figure 7 ) and with normal tissue perfusion (Figure 8 ) for a 30 volt constant voltage source ablation at 1, 3, 5, 10 and 15 minutes. The data demonstrate that the shapes of the temperature isotherms do not correlate well with tissue damage profiles. Tissue perfusion greatly affects the size of the resulting ablated region. At 15 minutes, lesion volumes are 267% larger without perfusion than with tissue perfusion. Figures 9 and 10 show a comparison of the temperature distribution and lesion size development with no tissue perfusion (Figure 9 ) and with normal tissue perfusion (Figure 10 ) at 1, 3, 5, 10, and 15 minutes following a 15 minute constant 30 volt ablation. In the case of no tissue perfusion, the lesion size continues to grow 14% within the first 5 minutes after radiofrequency energy is terminated. The lack of tissue perfusion prolongs the time needed to conduct the heat away from tissues near the surface of the ablation electrode. For cases with normal tissue perfusion, heat is quickly dissipated by tissue perfusion causing the lesion volume to stabilize in less than 2 minutes. By definition, the area of coagulative necrosis has no tissue perfusion. This accounts for the residual heating pattern within the ablated region as seen up to 3 minutes following the ablation. Figure 7 Comparison of temperature and lesion size development with no tissue perfusion for a 30 volt constant voltage source ablation. Ablation simulation results for a 30 volt constant voltage source ablation with no tissue perfusion. The results on the top half of the figure represent the temperature distribution surrounding the ablation probe in degrees Celsius. The results on the bottom half of the figure represent the percent tissue damage. The numbers listed at the bottom are the lesion volume sizes computed from the 63% cell damage isocontours at each time interval shown. Figure 8 Comparison of temperature and lesion size development with normal tissue perfusion (6.4 × 10 -3 m b 3 /m t 3 /s) for a 30 volt constant voltage source ablation. Ablation simulation results for a 30 volt constant voltage source ablation with normal tissue perfusion (6.4 × 10 -3 m b 3 /m t 3 /s). The results on the top half of the figure represent the temperature distribution surrounding the ablation probe in degrees Celsius. The results on the bottom half of the figure represent the percent tissue damage. The numbers listed at the bottom are the lesion volume sizes computed from the 63% cell damage isocontours at each time interval shown. Figure 9 Comparison of temperature and lesion size development post ablation with no tissue perfusion for a 30 volt constant voltage source ablation. Ablation simulation results following a 15 minute ablation without perfusion for a 30 volt constant voltage source ablation. The results on the top half of the figure represent the temperature distribution surrounding the ablation probe in degrees Celsius. The results on the bottom half of the figure represent the percent tissue damage. The numbers listed at the bottom are the lesion volume sizes computed from the 63% cell damage isocontours at each time interval shown. Figure 10 Comparison of temperature and lesion size development post ablation with normal tissue perfusion (6.4 × 10 -3 m b 3 /m t 3 /s) for a 30 volt constant voltage source ablation. Ablation simulation results following a 15 minute ablation with normal tissue perfusion (6.4 × 10 -3 m b 3 /m t 3 /s) for a 30 volt constant voltage source ablation. The results on the top half of the figure represent the temperature distribution surrounding the ablation probe in degrees Celsius. The results on the bottom half of the figure represent the percent tissue damage. The numbers listed at the bottom are the lesion volume sizes computed from the 63% cell damage isocontours at each time interval shown. A comparison of lesion volumes with no tissue perfusion computed using 63% and 100% iso-damage threshold contours and 42°C, 47°C, 60°C, and 90°C isothermal contours is presented for the cases of no tissue perfusion (Table 4 ) and normal tissue perfusion (Table 5 ). The sensitivity of the cell damage function (Figure 4 ) results in less than 10% differences in the size of lesions calculated using tissue damage thresholds of 63% and 100% cell damage. In contrast, volume sizes based on isothermal contours varies considerably at each temperature. When using traditional isothermal contours of 42°C and 47°C, the calculated lesion volumes are grossly overestimated by 500% and 167%, respectively. In both the case of no tissue perfusion and normal tissue perfusion, the 60°C isothermal contour resembles the lesion sizes calculated using the iso-damage contours. Table 4 Lesion Volume with No Tissue Perfusion Values represent the total volume of tissue necroses calculated over the course of the simulated ablation using various cell damage thresholds (D) and isothermal temperatures (IT). Source Voltage (Volts) D = 63% (mm 3 ) D = 100% (mm 3 ) IT = 42°C (mm 3 ) 1 IT = 47°C (mm 3 ) 1 IT = 60°C (mm 3 ) IT = 90°C (mm 3 ) 2.5 0 0 0 0 0 0 5.0 0 0 0 0 0 0 7.5 0 0 0 0 0 0 10.0 0 0 89 0 0 0 12.5 0 0 444 13 0 0 15.0 0 0 942 216 0 0 17.5 9 3 1649 526 6 0 20.0 121 67 2508 915 87 0 22.5 314 242 3549 1414 296 0 25.0 577 495 4860 2070 547 6 27.5 923 809 6452 2866 870 55 30.0 1388 1228 8386 3830 1243 202 1 – The 42°C and 47°C isothermal volumes were chosen specifically because they are frequently used to establish damage thresholds in hyperthermia and radiofrequency ablation, respectively. Table 5 Lesion Volume with 100% Normal Tissue Perfusion. Values represent the total volume of tissue necroses calculated over the course of the simulated ablation using various cell damage thresholds (D) and isothermal temperatures (IT). Source Voltage (Volts) D = 63% (mm 3 ) D = 100% (mm 3 ) IT = 42°C (mm 3 ) 1 IT = 47°C (mm 3 ) 1 IT = 60°C (mm 3 ) IT = 90°C (mm 3 ) 2.5 0 0 0 0 0 0 5.0 0 0 0 0 0 0 7.5 0 0 0 0 0 0 10.0 0 0 1 0 0 0 12.5 0 0 47 0 0 0 15.0 0 0 185 6 0 0 17.5 0 0 358 68 0 0 20.0 4 2 513 177 1 0 22.5 23 19 784 287 15 0 25.0 131 89 1068 488 107 0 27.5 251 222 1502 759 241 3 30.0 433 364 2014 1064 423 24 1 – The 42°C and 47°C isothermal volumes were chosen specifically because they are frequently used to establish damage thresholds in hyperthermia and radiofrequency ablation, respectively. Table 6 shows a comparison of lesion width and depth computed using 63% and 100% iso-damage threshold contours and 42°C, 47°C, 60°C, and 90°C isothermal contours. The data show overestimations of 30–77% in the width and 18–54% in the depth of lesions when using traditional isothermal temperatures for a 15 minute ablation with no perfusion. Table 7 shows that in cases with normal tissue perfusion, calculations using traditional isothermal contours results in overestimations of 25–88% in the width and 15–41% in the depth of lesions. Table 6 Lesion Dimensions with no Tissue Perfusion Values represent the maximum lesion width and depth calculated over the course of the simulated ablation using various cell damage thresholds (D) and isothermal temperatures (IT). Width (mm) Depth (mm) Source Voltage (Volts) D = 63% D = 100% IT = 42°C 1 IT = 47°C 1 IT = 60°C IT = 90°C D = 63% D = 100% IT = 42°C 1 IT = 47°C 1 IT = 60°C IT = 90°C 2.5 0 0 0 0 0 0 0 0 0 0 0 0 5.0 0 0 0 0 0 0 0 0 0 0 0 0 7.5 0 0 0 0 0 0 0 0 0 0 0 0 10.0 0 0 8 0 0 0 0 0 19 0 0 0 12.5 0 0 16 0 0 0 0 0 26 0 0 0 15.0 0 0 22 8 0 0 0 0 31 13 0 0 17.5 6 4 28 14 4 0 7 5 35 24 6 0 20.0 10 8 32 18 10 0 21 16 39 28 18 0 22.5 14 12 36 22 14 0 25 24 43 31 24 0 25.0 18 18 40 26 18 4 28 26 45 34 28 6 27.5 22 22 44 30 22 8 31 30 49 37 30 14 30.0 26 26 46 34 26 12 33 33 51 39 33 23 1 – The 42°C and 47°C isothermal volumes were chosen specifically because they are frequently used to establish damage thresholds in hyperthermia and radiofrequency ablation, respectively. Table 7 Lesion Volume with 100% Normal Tissue Perfusion Values represent the maximum lesion width and depth calculated over the course of the simulated ablation using various cell damage thresholds (D) and isothermal temperatures (IT). Width Depth Source Voltage (Volts) D = 63% D = 100% IT = 42°C 1 IT = 47°C 1 IT = 60°C IT = 90°C D = 63% D = 100% IT = 42°C 1 IT = 47°C 1 IT = 60°C IT = 90°C 2.5 0 0 0 0 0 0 0 0 0 0 0 0 5.0 0 0 0 0 0 0 0 0 0 0 0 0 7.5 0 0 0 0 0 0 0 0 0 0 0 0 10.0 0 0 2 0 0 0 0 0 2 0 0 0 12.5 0 0 8 0 0 0 0 0 16 0 0 0 15.0 0 0 10 0 0 0 0 0 25 0 0 0 17.5 0 0 14 6 0 0 0 0 27 7 0 0 20.0 4 4 16 8 2 0 5 4 29 23 2 0 22.5 6 6 20 10 6 0 10 8 31 25 8 0 25.0 10 8 24 14 8 0 23 23 33 27 23 0 27.5 12 12 26 18 12 4 25 25 36 29 25 5 30.0 16 14 30 20 16 6 27 27 38 31 27 9 1 – The 42°C and 47°C isothermal volumes were chosen specifically because they are frequently used to establish damage thresholds in hyperthermia and radiofrequency ablation, respectively. To validate the computational model, ablation experiments were performed at room temperature (22°C) in excised porcine liver tissue using 20, 25, and 30 volt constant voltage radiofrequency sources (500 kHz). Ablations were made for a 15 minute exposure time. Figure 11 shows that no visible lesion can be seen in tissues where the 20 volt constant voltage ablation was performed, as predicted by the computational simulation. A lesion that was approximately 10 millimeters in width and 22 mm in depth resulted from the 30 volt constant voltage ablation. Table 8 shows a high correlation between the computational data calculated at 22°C and the experimental results. Figure 11 Experimental validation radiofrequency ablation lesions in excised porcine liver tissue produced by a 20 volt (left) and 30 volt (right) constant voltage radiofrequency generator (500 kHz) for a 15 minute exposure time. The lesions were produced using a single needle ablation probe with a 2 centimeter uninsulated tip. Table 8 Comparison of Computational Data to Experimental Validation Data at 22°C. Lesion Width Lesion Depth Source Voltage (Volts) D = 63% (mm) Experimental (mm) D = 63% (mm) Experimental (mm) 20.0 0 0 0 0 22.5 0 -- 0 -- 25.0 4 5 5 6 27.5 8 -- 9 -- 30.0 10 10 21 22 Simulated lesion dimensions with no tissue perfusion at an ambient temperature of 22°C using a 63% cell damage threshold (D) were compared to experimental measurements made at 20, 25, and 30 volts using an experimental constant voltage radiofrequency ablation source. Lesions were generated with a single needle ablation probe with a 2.0 cm active electrode. All lesion measurements were measured visually under a micrsope at 10× magnification. Discussion To date, several computational studies have been performed to described the rate of lesion growth in radiofrequency ablation applications. In many cases, these studies use surrogate endpoints such as temperature isotherms and thermal dosing to calculate equivalent expressions for lesion size. While many models exists that account for far-more elaborate parameters such as tissue perfusion through large blood vessels, the interpretation of such models is difficult since most do not account for transient changes in tissue properties and often report tissue temperature only [ 6 - 9 , 12 , 14 - 19 , 56 , 57 , 60 ]. Several studies have identified that both exposure time and temperature contribute to tissue damage, however, few have actually calculated tissue damage. Those that do, have not allowed tissue damage to transiently influence the electrical and thermal properties of tissues [ 6 , 56 ]. In this study, we created a computational simulation that tested some of the basic assumptions made in modeling lesion growth problems. We developed a model where tissue perfusion and the electrical conductivity are allowed to vary at each time step and spatial position as a function of tissue damage and temperature. These simulations are significantly more time-consuming since gross simplifications to heating mechanisms are not made. Although our model geometry is simpler than others that appear in the literature, we chose to ignore large vessels since their position and impact are highly variable. We chose a simpler geometry so that the impact of damage-dependent tissue perfusion and temperature-dependent electrical conductivity could be assessed more directly. The damage-dependent tissue perfusion accounts for physiological observations of tissue coagulation and local cessation of blood flow. Unlike thermal dosing, where thermal injury is calculated globally over the entire duration of an ablation, tissue damage is calculated at every time step. The intermediate tissue damage that results at every timestep influences the local tissue perfusion and creates a moving boundary condition which changes the local heat sink properties. Ignoring the intermediate timesteps causes tissue perfusion to remain constant throughout the entire ablation, which results in an underestimation of the true lesion size. The use of temperature-dependent electrical conductivity greatly affects modeling results, as the electrical conductivity has been shown to increase dramatically over the course of tissue heating [ 57 ]. When constant electrical conductivity is used, the SAR is grossly underestimate, which also results in an underestimation in lesion size. An important outcome of this study is the demonstration that, temperature isotherms and tissue damage patterns are not synonymous. Traditional use of temperature isotherms that are used to define lesion size rely on coagulation temperature for protein (42–47°C) and grossly overestimate lesion dimensions. Our studies show that temperature decrease is gradual, while tissue damage decreases rapidly as a function of distance. It is this sharp decrease in tissue damage that causes lesion boundaries to appear fuzzy, as predicted by our model. The results also demonstrate that ablation lesions continue to grow after the applied power is terminated. Lesions continue to grow while temperature envelopes collapse after ablation since sufficiently high temperature are present to accrue tissue damage. In nearly all cases, lesions continued to grow several minutes following the ablation. A comparison of the resulting lesion dimensions between fully perfused and non-perfused tissues show that the lesion width decreases 38–46% and the lesion depth decreases 18–20% when tissue perfusion is accounted for in the model. Previous studies have shown that tissue perfusion can account for as much as 50% change in the size of the lesions generated during ablation [ 59 ]. An important observation in this study is the resemblance of the 60°C isocontour to lesion size. While the 42°C and the 47°C isotemperature contours are poor indicators of lesion size, 60°C is highly correlated with the lesion volumes. Seemingly, this would suggest that time-intensive tissue damage calculations need not be made since a critical temperature of 60°C can be used to identify lesion size. However, this is only true if the calculated temperature is a function of both transient changes in tissue perfusion and the electrical conductivity. In the absence of either of these phenomena, lesion sizes calculated at 60°C would underestimate lesion size. The validation data demonstrate that the model accurately accounts for the behavior of lesion growth in tissue. There are, however, a few limitations to this model. First, it is well established that temperature elevation of tissues results in the denaturing of proteins, which may drastically change the electrical conductivity of tissue in a nonlinear fashion [ 51 , 57 ]. Preliminary data suggests that the electrical conductivity substantially increases, which would likely increase the rate of tissue damage. The results of this study show that as a first order approximation the conductivity of equivalent sodium chloride solutions produces results that are within 5% of the experimental measurements. Although the phenomena described in this reporting are applicable to different tissues, the resulting lesion dimensions and temperature profiles in this study apply only to liver tissue. Similar studies can be made in other tissues, but were not pursued in this study. A second limitation in our model is that it is only valid for temperatures below 100°C. At temperatures above 100°C, tissues begin to boil and generate gas. When this occurs, some of the energy that contributes to temperature increase is used to change the water content of tissues into gas. At substantially higher temperatures, the composition of gas may be highly complex as tissue begins to burn and break down. Although gas generation is commonly seen in clinical use of radiofrequency ablation, impedance rises due to tissue charring limit the progressive rise in temperature. The complexity of multi-phasic ablation was beyond the scope of this study. Disclaimer The mention of commercial products, their sources, or their use in connection with material reported herein is not to be construed as either an actual or implied endorsement of such products by the Department of Health and Human Services.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC514567.xml
533872	The use of Open Reading frame ESTs (ORESTES) for analysis of the honey bee transcriptome	Background The ongoing efforts to sequence the honey bee genome require additional initiatives to define its transcriptome. Towards this end, we employed the Open Reading frame ESTs (ORESTES) strategy to generate profiles for the life cycle of Apis mellifera workers. Results Of the 5,021 ORESTES, 35.2% matched with previously deposited Apis ESTs. The analysis of the remaining sequences defined a set of putative orthologs whose majority had their best-match hits with Anopheles and Drosophila genes. CAP3 assembly of the Apis ORESTES with the already existing 15,500 Apis ESTs generated 3,408 contigs. BLASTX comparison of these contigs with protein sets of organisms representing distinct phylogenetic clades revealed a total of 1,629 contigs that Apis mellifera shares with different taxa. Most (41%) represent genes that are in common to all taxa, another 21% are shared between metazoans (Bilateria), and 16% are shared only within the Insecta clade. A set of 23 putative genes presented a best match with human genes, many of which encode factors related to cell signaling/signal transduction. 1,779 contigs (52%) did not match any known sequence. Applying a correction factor deduced from a parallel analysis performed with Drosophila melanogaster ORESTES, we estimate that approximately half of these no-match ESTs contigs (22%) should represent Apis -specific genes. Conclusions The versatile and cost-efficient ORESTES approach produced minilibraries for honey bee life cycle stages. Such information on central gene regions contributes to genome annotation and also lends itself to cross-transcriptome comparisons to reveal evolutionary trends in insect genomes.	Background The honey bee, Apis mellifera , occupies a prominent place in biological research due to its social behavior, learning capabilities, haplodiploid mechanism of sex determination, and plasticity in phenotype (caste) and longevity. Thus, it is a model organism for classical and sociogenetic studies. In addition, bees drive a large-scale apicultural industry, and also generate important income in small-scale subsistence beekeeping. And finally, bees are of great economic and ecological relevance for their role as generalist pollinators. The decision to include the honey bee amongst the current organisms for complete genome sequencing, was, therefore, well founded, yet information on its transcriptome is still meager. When starting this study, little over 250 genes were annotated as partial or full length coding sequences, and only about 15,500 expressed sequence tags (mainly 5'-ESTs generated from a normalized bee brain cDNA library [ 1 ]) were available in public databases. Thus, even after sequencing of the honey bee genome will be completed a considerable transcriptome sequencing effort will still be required for unequivocal genome annotation, gene identification, and subsequent functional studies. We used the ORESTES (Open Reading frame Expressed Sequence Tags) strategy to generate ESTs from different life cycle stages of the honey bee, such as appropriate for a genome annotation initiative. This strategy preferentially generates ESTs of the central, and thus most informative portion of the transcript [ 2 ], and frequently also identifies less abundant mRNAs [ 3 ]. The efficacy of the Open Reading frame ESTs strategy, in the context of an organism for which there is limited genomic information, has recently been demonstrated for Schistosoma mansoni [ 4 ]. This cost-efficient approach increased the already existent Apis EST database by 30% new reads. Of the 5,021 ORESTES, only 35.2% matched with previously deposited Apis ESTs. When assembled with the existent Apis ESTs in the NCBI database, the ORESTES sequences extended 66% of the mixed contigs. Together these data indicate that the ORESTES methodology could effectively complement the current efforts towards the definition of the Apis transcriptome. Results and discussion Honey bee Open Reading frame ESTs We generated a total of 87 mini-libraries from the four major life cycle stages of honey bee workers (embryo, larva, pupa, adult) by the use of arbitrary primers and a low-stringency RT-PCR protocol [ 2 ]. From these libraries we obtained 5,021 sequences of appropriate standard quality (sequence > 100 bases; Phred 15) and with an average size of 373.9 bp. These sequences were deposited in the GenBank EST database (accession numbers CK628548 to CK633568). In the annotation pipeline, these were first submitted to BLASTN searches against Apis mellifera sequences deposited in the NCBI EST database (dbEST). At this step, 35.2% (1,769) of the validated sequences matched Apis ESTs (Table 1 ). In a subsequent step, a BLASTX comparison of the remaining sequences against the nr-NCBI database permitted the annotation of an additional 22.4% (1,123) of the honey bee ORESTES, while the remaining 42.4% (2,129) did not match any known sequence. This rather large set of ESTs that did not result in significant alignment with any sequence deposited in non-redundant databases contains candidates for novel honey bee genes. Table 1 Apis mellifera Open Reading frame ESTs. Sequencing results Number of reads Total analyzed reads 5021 - Embryos 1358 - Larval stages 720 - Pupae 1219 - Adults 1479 - Stage mix 245 Local alignment matches - Apis ESTs from dbEST 1769 a - Apis mellifera sequences in GenBank 16* b - Genes of other organisms (orthologs) 1123 b - No matches in GenBank 2129 Clusterization results - Number of contigs 488 - Number of singlets 893 - Total number of clusters 1381 a BLASTN against dbEST using an E-score of 10 -30 as cutoff value; b BLASTX against the nr database used <10 -15 as the cutoff. *This set is also represented by ESTs in the dbEST database. It is included here as additional information only and is not to be summed up with the other matches. The 5,021 Apis ORESTES were assembled by CAP3 into 488 contigs of a mean size of 519 bp, leaving 893 singlets. In a second round of BLASTX comparisons against the nr-NCBI database, 28.5% of the contigs and 9.2% of the singlets were classified as putative orthologs. When the respective best matches were classified according to species or higher order taxa (Figure 1 ), 89.6% were from the arthropod clade (including fully or partially sequenced Apis mellifera genes). The largest fraction of these putative orthologs showed best matches with predicted Anopheles genes (43.9%), followed by ORESTES that were classified as putative orthologs of Drosophila (29.5%). Figure 1 Distribution of Best-BLASTX-matches for assembled Apis mellifera Open Reading frame ESTs. After assembly into contigs and singlets the sequences were submitted to a search against a non-redundant protein database (NCBI). Independent of its E-score, the best match in each BLASTX result was listed according to organism category. Gene Ontology classification We assigned level 3 Gene Ontology (GO) classifications to 326 of the total of 488 assembled contigs; 162 contigs did not match any sequence in the nr-protein database. In the manual annotation preceding the GO analysis we preferentially assigned the contigs with respect to their Drosophila orthologs. The cellular component, biological process, and molecular function classifications of the honey bee sequences are shown in Table 2 . In the biological process categories there is a clear prevalence for ESTs representing cell communication, cell growth and maintenance, metabolism and morphogenesis. For molecular function, the dominant assignments were to enzymatic activity and to nucleic acid binding and related functions (translation factor, transcription factor). When compared to the corresponding GO results obtained for the bee brain ESTs [ 1 ], we noted a similar distribution in category dominance structure, except for the molecular functions 'transporter and ligand binding/carrier' which have a higher representation in the bee brain ESTs than in our ORESTES contigs. This discrepancy most probably reflects functional differences in the tissues used in these two studies. Table 2 Gene Ontology classification of Apis mellifera ORESTES contigs according to the Drosophila genes that they represent. Gene Ontology Number of genes Cellular Component extracellular matrix 4 extracellular space 5 intracellular 99 membrane 29 others 8 Biological Process reproduction 18 cell motility 7 response to stress 6 cell communication 25 pattern specification 10 cell growth and/or maintenance 55 metabolism 79 response to external stimulus 10 morphogenesis 24 embryonic development 9 cell differentiation 9 others 41 Molecular Function nucleotide binding 14 nucleic acid binding 40 RNA polymerase II transcription factor activity 7 antimicrobial peptide activity 3 helicase activity 4 receptor signaling protein activity 5 structural constituent of cytoskeleton 5 microfilament motor activity 5 transcription factor activity 6 kinase activity 14 oxidoreductase activity 22 transferase activity 23 hydrolase activity 38 protein binding 29 metal ion binding 9 ion transporter activity 8 others 70 GO levels were set at 3. In either of the GO categories, individual contigs may be listed in more than one category. This GO classification only includes Apis mellifera orthologs to Drosophila genes that are represented by a Flybase code. Clustering of honey bee ESTs We clustered the contigs generated in this study (AmORESTES contigs) with the Apis mellifera ESTs already present in the NCBI dbEST database (further referred to as AmNCBI contigs). Clustering performed by CAP3 resulted in a total of 3,408 contigs and led to a general increase in read depth (Figure 2A ). This increase in read depth is reflected in the CAP3 assembled mixed sequences of the two databases. Mean length is 696 bp for the AmNCBI contigs and 496 bp for the AmORESTES contigs (Figure 2B ). For the mixed contigs we noted a mean increase of about 150 bp in contig length, thus documenting that the ORESTES sequences add considerable information to the characterization of the honey bee transcriptome and for subsequent studies of specific genes. Figure 2 CAP3 assembly of Apis mellifera Open Reading frame ESTs (AmORESTES) with Apis mellifera ESTs previously deposited in dbEST (AmNCBI). A) Read depth distribution of pure AmNCBI or AmORESTES and of mixed contigs; B) EST size distribution of these contigs, C) Details of individual mixed contigs showing the extension and gap-closing characteristics. In all graphs, AmORESTES sequences are in blue, AmNCBI contigs are in red, and mixed contigs are in green. Within the total contig population, 9.5% of the assembled sequences (323) are represented by mixed contigs of both AmORESTES and AmNCBI sequences, and within this group 66.3% (214) of the original AmNCBI contigs were considerably extended, or were joined across gaps by the AmORESTES contigs, as illustrated in Figure 2C . The fact that the number of mixed contigs is relatively low compared to total contig number may be attributed to two aspects. First, most of the AmNCBI contigs were obtained from a single tissue (brain) library, whereas the AmORESTES sequences represent whole body transcripts of all life cycle stages of the honey bee. Second, the AmNCBI sequences are mainly 5'-ESTs, whereas the AmORESTES sequences are expected to cover more central cDNA regions. Genome comparison Even though the total number of ESTs available for Apis mellifera is still low when compared to established genomic model organisms, we performed an across genome analysis with the set of 3,408 honey bee contigs. This involved sequential BLASTX searches, using the honey bee sequences as query entries against protein databases of Drosophila melanogaster, Anopheles gambiae, Caenorhabditis elegans , human, protozoan and fungal origin. With this selection of organisms we intended to extract information on the percentage of genes that Apis shares (i) with all organisms, (ii) with animals, (iii) with different sets of metazoans, (iv) and exclusively with insects. The cutoff E-value in these comparisons was set at 10 -6 , as used in comparisons of similar nature [ 4 ], and the representation of the respective putative orthologs was listed across taxonomic levels. We found that 1,629 Apis contigs presented significant match with sequences belonging to at least one of the taxa genomes. From these, 460 contigs (28.2%) correspond to genes with a representation in all the above taxa (Figure 3 ). In addition, further 211 contigs (12.9%) could also be classified as common to all organisms since they were represented in all but in one of the members of this set of taxa (at this level, Anopheles and Drosophila were considered as a single group representing Diptera). This increases the set of EST contigs that the honey bee may share with all organisms to 41.2%, or, when considering the entire set of 3,408 contigs, to 19.7%. The second largest set of ESTs (312 + 37 contigs) is the one that is represented as genes common to the bilaterian clade (or metazoans in general), and only the third largest set (198 + 68 ESTs) contains genes that are represented solely in hymenopterans and dipterans, and thus in the insect clade. Figure 3 Similarity and representation pattern of assembled Apis mellifera ESTs (ORESTES + NCBI dbESTs) with predicted proteins of other organisms. In this comparison we included eukaryotes with completely sequenced genomes ( Drosophila melanogaster , Anopheles gambiae , Caenorhabditis elegans and human), plus higher taxon groups, such as protozoans (primarily represented by Plasmodium falciparum and P. yoelii ) and fungi (primarily represented by Saccharomyces cerevisiae , Schizosaccharomyces pombe and Neurospora crassa ). These BLASTX comparisons were performed with an E-value cut-off level set at 10 -6 . Subsequently, the representation pattern of each of the Apis ESTs in each of the eukaryotic genomes was listed. Out of the total 3,408 Apis EST contigs, 1,629 could be classified as putative orthologs, and these were grouped according to the representation of these genes at the different taxonomic levels. Since deep-level phylogeny relationships within the bilateria are still a matter of debate, we separated our dataset according to the two prevalent hypotheses. The traditional view clusters arthropods within the coelomate clade. In our set of genomes, this tree architecture would be represented by genes shared between insects and the human genome. The alternative, more recently proposed hypothesis joins arthropods with nematodes to form an ecdysozoan clade [ 5 ]. The result of our comparison, which places emphasis on shared genes and not on the frequency of gene losses, is more consistent with the traditional view, since the coelomate clade is represented in this analysis with almost five times more shared genes than the ecdysozoan clade. To infer on functional aspects within this pattern of genes that different clades appear to have in common we performed a Gene Ontology classification on biological process. In the set of Apis ESTs that stands for genes putatively shared with all organisms, the majority was classified as having a role in metabolism, and thus can be considered to represent basic functions. In contrast, the majority of Apis ESTs that are shared within the insect clade was represented in the biological process categories of cell growth and/or maintenance and cell communication. The corresponding insect-specific genes are therefore supposedly involved in more specialized functions. A similar conclusion can be reached from the micro- and macroarray analyses of transcripts detected in adult honey bee workers performing different tasks during their adult life cycle [ 6 , 7 ]. A total of 70 putative ortholog ESTs did not comply with any of the plausible phylogenies, yet nevertheless, this set may contain ESTs of interesting information content, especially when considering that the main set of genes within this group consists of Apis mellifera contigs that overlap with a mammalian genome. A manual analysis of these 23 contigs by BLASTX against the nr-NCBI database revealed that they are (at least by three orders of magnitude in E-values) more similar to mammals (especially to humans) than to other vertebrates and even other insects. This suggests that these genes may have diverged less in Apis and mammals and, therefore, may be subject to related selection pressure. Alternatively, at least some of them may have been modified in Diptera, and thus would show up as insect genes only once further non-dipteran insect genomes or transcriptomes have been sequenced and annotated. As shown in Table 3 , this set of bee/human contigs contains a considerable number of predicted proteins related to cell signaling/signal transduction and transcription factors. Such a bias to information processing in our dataset of genes shared between honey bees and a mammalian genome may reveal system properties related to complex functions. Table 3 Annotation and Gene Ontology characteristics of 11 honey bee EST contigs sharing significant similarity with mammalian but not with other vertebrate or invertebrate sequences. In all cases, the best match was with human proteins. For these 11 out of 23 contigs we could retrieve functional information. Apis EST contig E-score value a GO, Biological process b Human LOCUS ID GenBank annotation Additional information Human Insect 437 2e-67 NA without result NM_019116: ubiquitin binding protein ubiquitin-specific protease domain 663 3e-10 NA without result NM_182830: MAM domain contactin 5; neural adhesion molecule 1081 5e-07 9.7 without result NM_013041: RAB3A interacting protein (rabin3)-like1 guanin nucleotide exchange factor domain 1425 8e-27 NA without result NM_182565: hypothetical protein MGC29814 TBP-associated factor 4; TATA box binding protein 1674 4e-82 NA without result NM_172374: interleukin 4 induced 1 none 1953 7e-11 NA without result NM_024707: gem (nuclear organelle) associated protein spliceosomal snRNP biogenesis 2807 8e-07 5e-04 regulation of physiological process NM_138457: forkhead box P4 transcription factor activity 2896 4e-05 0.002 reproduction, metabolism NM_004654: ubiquitin-specific protease 9 ubiquitin thiolesterase activity 3167 2e-27 7e-05 cell communication NM_033046: rhotekin signal transduction 3347 5e-26 7.5 without result NM_014006: PI-3-kinase-related kinase SMG-1 involved in nonsense-mediated mRNA decay 3374 1e-10 0.18 cell communication, NM_014035: sorting nexing 24 intracellular signaling cascade a E-value of the contig alignment with human or known insect sequences, NA: did not show any alignment; b Gene Ontology results on Biological Process, FatiGO level 3, c additional information obtained from Entrez Gene – NCBI and GOA link (GOAnnotations@EBI – European Bioinformatics Institute). Finally, we found that 1,779 (52,2%) of the assembled EST contigs did not match with any sequence of the analyzed organisms. Such a large proportion of Apis- specific contigs is likely to be an overestimate. As noted in a previous study [ 1 ], this might be partly due to technical problems, such as, sequencing of cDNA inserts consisting mainly of 3'-untranslated regions, the presence of unspliced intron sequences, cDNAs with a negative reading frame, or chimaeric cDNAs. However, the major portion of the Apis -specific contigs may have become classified as species-specific due to their relatively short ORFs. We performed an ESTScan analysis on the Apis -specific contigs which detected ORFs in 56% of the assembled ESTs. These ORFs are, however, relatively short, with a mean ORF length around 280 bp. Short ORF length represents a notorious problem to alignment algorithms resulting in low match scores, and consequently, a more frequent classification of short ORF ESTs as species-specific transcripts. For the honey bee, this has been shown for the brain cDNA library where 84% of the ESTs with ORFs shorter than 450 bp were classified as species-specific, against 24% in the EST set that had ORFs larger than 450 bp [ 1 ]. In order to gain a general perspective on the representation of species-specific ESTs we also directed our attention to estimates obtained in whole-genome cross-species analyses. For instance, a figure of 18.6% of species-specific genes was ascertained for Drosophila melanogaster in a genome comparison which included Anopheles gambiae as the other insect representative [ 8 ]. Based on this information, and taking advantage of a set of Drosophila melanogaster ORESTES, generated in a parallel project, we calculated the frequency of Drosophila -specific ORESTES sequences to obtain a more realistic estimate on Apis -specific genes in relatively small sets of ESTs. In this analysis, a set of 5,000 CAP3 assembled Drosophila ORESTES (409 contigs) was submitted to sequential BLASTX searches against protein databases of Drosophila melanogaster , Anopheles gambiae , Caenorhabditis elegans , human, protozoan and fungal, as described for the Apis contigs. For comparison, this same analysis was also performed with Apis ESTs, using separately 5,000 AmORESTES (486 contigs) and 5,000 AmNCB (632 contigs). The Drosophila and Apis EST contigs consistently showed relatively low proportions of insect-specific genes (6–13%). Still lower (ca. 1% each) was the proportion of ESTs that had significantly higher similarity scores with eukaryotes other than Insecta. In all EST sets we found a large fraction of sequences that were classified as either Drosophila -specific (40%) or Apis -specific (51% for AmNCBI dbESTs and 47% for AmORESTES). This high proportion of species-specific genes, therefore appears to be generated independent of the method used in EST sequencing, as it is represented in similar proportions in both the ORESTES set and the conventional 5'-EST set (Figure 4 ). Figure 4 Percentage of honey bee and Drosophila ESTs representing putative species-specific genes (blue bars) in relation to ESTs that represent genes solely shared within the insect clade (pink bars), or that have higher similarity with eukaryotes other than the insect clade (yellow bars). In separate comparisons, the Apis mellifera contigs (ORESTES + NCBI dbESTs, n = 5,000), AmORESTES (n = 5,000), AmNCBI dbESTs (n = 5,000), and Drosophila melanogaster ORESTES contigs (n = 5,000) were analyzed against protein databases of an insect ( Anopheles gambiae ) and several non-insect species ( C. elegans , protozoans, fungi and H. sapiens ) with completely sequenced genomes. The cut-off E-value in these comparisons was set at 10 -6 . The figure of 40% Drosophila -specific genes obtained for our Drosophila ORESTES set can be directly set in contrast with the estimate of 18,6% species-specific genes reported in the Drosophila genome based study [ 8 ], and this would predict an overestimate factor of 2.15 for species-specific genes in the EST sets. When this factor is applied to the honey bee ORESTES, the 47% estimate for species-specific AmORESTES can thus be corrected to a more realistic figure of 22%. This estimate is in agreement with the results of Whitfield et al. [ 1 ] who observed that 24% of the honey bee genes represented by ESTs with ORFs larger than 450 bp did not have matches to any known protein sequences. This Apis -specific gene estimate is also in range when considering that the two dipteran species are thought to have separated from a common ancestor approximately 250 million years ago, whereas the postulated sister-group relationship of Hymenoptera and Mecopteroidea [ 9 ] suggests a pre-permian divergence, with a predicted separate lineage evolution of over 280 million years for honey bees and dipterans [ 10 ]. Conclusions The generation of a relative small set of Open Reading frame ESTs (ORESTES) that match and complement the already existent Apis EST database shows that this approach is sufficiently robust and favorably complements other strategies, such as ESTs prepared from normalized cDNA libraries. Its inherent properties of detecting transcripts of low abundance and aligning with central regions of transcripts [ 2 , 3 ] also make it a suitable tool in searches for novel honey bee genes and their annotation in parallel with ongoing genome sequencing projects. Furthermore, the genome comparisons performed in this and other studies [ 1 , 11 ] highlight that the elevated number of putative Apis -specific genes will still require extensive transcriptome sequencing for high quality genome annotation, and will play an important role in the question of insect genome organization and model systems in comparative studies [ 12 ]. Methods Biological samples and RNA extraction Samples of the four major stages of the honey bee life cycle were collected from Apis mellifera colonies (Africanized hybrids) kept in the experimental apiary of the Dept. Genetics, Univ. São Paulo, Campus Ribeirão Preto, Brazil. Each embryo sample contained approximately 300 eggs retrieved from a frame on which the queen had been caged for up to 72 hours. This assured that we covered the entire embryonic period. The larval sample was a representation of all five instars and included also spinning-stage larvae. Prepupae and pupae, including white-eyed, pink-eyed, brown-eyed and pigmenting pupae, were pooled into the pupal samples. For the adult sample we collected newly emerged bees, a random sample of hive bees (picked from a brood frame), and returning foragers. All these samples were snap frozen in liquid nitrogen. Total RNA was isolated using TRIzol reagent (Invitrogen). The lipid-rich larval and pupal samples required two additional extraction steps with phenol/chloroform and chloroform to obtain RNA of adequate purity. In the case of Drosophila melanogaster , dechorionated embryos, larvae plus prepupae and pupae, as well as adult flies were collected from an isogenic y , w 1118 stock of Drosophila melanogaster . These were immediately frozen in liquid nitrogen and stored at -80°C until use. Total RNA was extracted with TRIzol, as described for Apis mellifera . Generation of Open Reading frame ESTs (ORESTES) From high quality DNA-free total RNA samples we isolated poly(A) + RNA using an Oligotex II (Qiagen) kit. To assess poly(A) + RNA quality of the samples we performed Northern blot hybridizations with an actin ( Apis mellifera ) or tubulin ( Drosophila melanogaster ) probe. The probes were labeled by a random priming reaction in the presence of [α- 32 P]dCTP. The actin fragment was amplified using the primers described in Table 4 . The Drosophila tubulin probe was already available from previous studies. High quality total RNA preparations were subjected to a DNase I treatment, and the absence of DNA contaminants was assessed by Southern blot hybridization of PCR products amplified with Apis or Drosophila 16S mitochondrial DNA primers, respectively. High quality poly(A) + RNAs were aliquoted and stored at -80°C. Table 4 Specific primers used to assess quality and absence of DNA contaminants of the RNA samples, and randomly selected primers used to generate cDNA profiles. Primer code Sequence actin F (Apis) 5' AGCTATGAACTTCCAGATGGT 3' actin R (Apis) 5' CCACATCTGTTGGAAGGT 3' 16S mitochondrial F (Apis) 5' TTATTCACCTGTTTATCAAAACAT 3' 16S mitochondrial R (Apis) 5' 'TATAGATAGAAACCAAYCTG 3' 16S mitochondrial F (Drosophila) 5' CCGGTCTGAACTCAGATCACGT 3' 16S mitochondrial R (Drosophila) 5' CGCCTGTTTAACAAAAACAT 3' p3_2 5' TTGGGGATCGTATGTAGTATG 3' pA82_1 5' CACTTCAGGATCCCTTGTAAGC 3' pA82_2 5' CCAACATTGAATTCTCTTTGAC 3' pA82_4 5' CAATAACAATGAATTCCAGAATCTCG 3' pPT7C4_B 5' GCTTACAAGGGATCCTGAAGTGTTTCC 3' pPT7C4_XS 5' GCAGGTAAACTCTACTCGAGTTACG 3' M-RON-AS 5' CCAGGATGTTTGGGTGATGTA 3' CREB-S 5' TCATGCAACATCATCTGCTCC 3' H-SPARC-S 5' CTAACCCAAGACATGACATTC 3' M-CD151-S 5' AAAGCTCGGAGGCAGCGAACT 3' H-CD151-AS 5' CATGTGGCTGCAAGGCAAAGC 3' M-SPARC-AS 5' GCCCAATTGCAGTTGAGTGAT 3' M-ETS1-AS 5' GTCTTGATGATGGTGAGAGTC 3' FUT-3-S 5' TCATGTCCAACCCTAAGTCAC 3' FUT-3-AS 5' TCCAGCAGGCCTTGCAGAAAT 3' M-CMET-S 5' TATCTCAAACGATCGAGAGAC 3' M-CMET-AS 5' GCACATCTATTACCAGCTTTG 3' H-CMET-S 5' TTTCAAATGGCCACGGGAC 3' H-CMET-AS 5' GCACATTTATGACCATTCTCG 3' H-Rhoc-AS 5' AGAAACAACTCCAGGGGCCTG 3' M-Rhoc-AS 5' CTACCCAAAGCAGAAACCCCA 3' H-Sparc-AS 5' CCAAAACCATCCTTGACAACA 3' H-RON-AS 5' TGATGAGGTCCTTCACGGTG 3' B237-2 5' CGGAATTCACCAGATTTGAACAGAAGAG 3' B237-3 5' AACTGCAGTTAACCAGATTTGAACAGAAA 3' GST_(PGEX)_NHE_I-S 5' CCGCTAGCATGTCCCCTATACTAGGTTA 3' HOXA_I-F 5' CGCTCCCGCTGTTTACTCT 3' P21-RasaI-F 5' GACCGCTCCTCCAACTAACC 3' P21-RasaI-R 5' CCGGCCCACCTCTTCTACTA 3' SRY8299.2 5' TCTCTTTATGGCAAGACTTACG 3' SRY1532.1 5' TCCTTAGCAACCATTAATCTGG 3' 92R7.2 5' GCCTATCTACTTCAGTGATTTCT 3' TAFIEX.1R 5' ATCCAAGGTTCTCCCAATA 3' ORESTES profiles were generated according to Dias-Neto et al. [ 2 ]. Briefly, aliquots of 15 ng of purified mRNA were subjected to reverse transcription reactions utilizing SuperScript II Reverse Transcriptase (Invitrogen) and a set of randomly selected primers (Table 4 ). First-strand cDNA synthesis occurred at 37°C for 60 min in a total volume of 20 μl. The products of this reaction were diluted 1:5 in water and stored at -20°C. The cDNAs contained in 1 μl of each diluted RT-product were then amplified by PCR using the same or a single alternative random primer in a PCR mix (Ready-to-Go PCR bead, Amersham Biosciences). The amplification protocol consisted of an initial step at 75°C for 5 min, followed by a 45 cycles touchdown series (95°C for 30 s, a gradually decreasing annealing temperature from 66 to 44°C lasting 10 s per step and a decrease of 2°C per step, 72°C for 1 min), and a final extension reaction at 72°C for 7 min. Aliquots of the PCR products (3–5 μl) were run on 1% agarose gels and stained with ethidium bromide. From profiles that presented near-even smears we excised two sets of amplification products, one covering a size range from 300 to 700 bp and a second one from 700 to 1500 bp. For cloning, these were extracted from the agarose gels (QIAquick Gel Extraction kit, Qiagen) and ligated into pUC18 (SureClone Ligation kit, Amersham Biosciences) for transformation of competent E. coli DH5α-cells by heat shock. Bacteria were grown in 2 × YT medium before aliquots were plated on 2 × YT agar containing ampicillin. Blue-white selected positive colonies were picked, grown overnight in 2 × YT medium in 96-well plates, and used as templates for PCR using vector primers (M13 forward and reverse). An aliquot of each amplification product was analyzed on a 1% agarose gel before another 1 μl aliquot was submitted to DNA sequencing using standard protocols of the DYEnamic™ ET Terminator kit (Amersham Biosciences). The reaction products were analyzed in a MegaBACE™ 1000 automated sequencer. Only profiles with more than 80% positive PCR reactions were sequenced. Sequence analysis After passing through the Base Caller Cimaron 1.53 Slim Phredfy (insert size > 100, "N" nucleotides less than 20%, and "N" repetitions of less than 6 nucleotides) and ScoreCard procedure (MegaBACE™) to check sequences quality, reads that were larger than 100 nt were submitted to an automated protocol for data analysis (Gene Annotation Pipeline) of the Apis mellifera or Drosophila melanogaster ORESTES. The protocol consisted of the following steps: conversion of electropherograms (Phred, to formats .fasta, .phd and .qual), primer and vector detection and trimming (Cross_match) and masking of repeats (RepeatMasker). Validated fasta format sequences were then submitted to a general BLASTN search against GenBank entries for mitochondrial and rRNA, as well as bacterial and fungal RNA to detect and eliminate contaminant sequences. For the Apis mellifera ORESTES, subsequent BLASTN searches were performed against the approximately 15,500 Apis mellifera EST sequences deposited in GenBank dbEST. In this case, significant E values were set at 10 -30 . Searches against the non-redundant protein database entries used the BLASTX option with E-values set at 10 -10 as significance cut-off level. CAP3 was used to clusterize the ORESTES sequences of both species. For Apis mellifera , the annotation of the 488 contigs was manually checked, giving preference to Drosophila sequences in the Unigene assignment. Subsequently, the contigs were batch submitted to a Gene Ontology procedure utilizing the FatiGO tools [ 13 ] . Clusterization of the Apis ORESTES contigs and singletons with the Apis mellifera ESTs deposited in GenBank dbEST was also performed using a CAP3 routine (standard parameters). Authors' contributions Apis ORESTES: FMFN and JFS participated in all steps of library preparations data and analyses; MAVC and DGP performed the bioinformatics analyses; RMM, PMVP and MFRS participated in the library preparations and GO analysis; MCRC sequenced libraries; AMN performed validation PCRs on selected ORESTES; AEE participated in the design of the study and preparation of biological material; MMGB, EME, FSE and ZLPS participated in the design of the study, library preparations and conceptual data analysis; MLPL, VV and KH participated in the design of the study, library preparations and prepared the manuscript; WASjr coordinated the design of the study and the bioinformatics analysis. Drosophila ORESTES: VV, JFS, DDA, RMM and EDN participated in all steps of library preparations and analyses; NM and RGRP participated in the design of the study and preparation of biological material; LFLR, WKM and AFC participated in RNA sample preparation; SJS, MAVC and WASjr participated in the design of the study and performed the bioinformatics analyses; AJGS, MAZ, EME and MLPL conceived and coordinated the study. All authors read and approved the final manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC533872.xml
546235	A finite element method model to simulate laser interstitial thermo therapy in anatomical inhomogeneous regions	Background Laser Interstitial ThermoTherapy (LITT) is a well established surgical method. The use of LITT is so far limited to homogeneous tissues, e.g. the liver. One of the reasons is the limited capability of existing treatment planning models to calculate accurately the damage zone. The treatment planning in inhomogeneous tissues, especially of regions near main vessels, poses still a challenge. In order to extend the application of LITT to a wider range of anatomical regions new simulation methods are needed. The model described with this article enables efficient simulation for predicting damaged tissue as a basis for a future laser-surgical planning system. Previously we described the dependency of the model on geometry. With the presented paper including two video files we focus on the methodological, physical and mathematical background of the model. Methods In contrast to previous simulation attempts, our model is based on finite element method (FEM). We propose the use of LITT, in sensitive areas such as the neck region to treat tumours in lymph node with dimensions of 0.5 cm – 2 cm in diameter near the carotid artery. Our model is based on calculations describing the light distribution using the diffusion approximation of the transport theory; the temperature rise using the bioheat equation, including the effect of microperfusion in tissue to determine the extent of thermal damage; and the dependency of thermal and optical properties on the temperature and the injury. Injury is estimated using a damage integral. To check our model we performed a first in vitro experiment on porcine muscle tissue. Results We performed the derivation of the geometry from 3D ultrasound data and show for this proposed geometry the energy distribution, the heat elevation, and the damage zone. Further on, we perform a comparison with the in-vitro experiment. The calculation shows an error of 5% in the x-axis parallel to the blood vessel. Conclusions The FEM technique proposed can overcome limitations of other methods and enables an efficient simulation for predicting the damage zone induced using LITT. Our calculations show clearly that major vessels would not be damaged. The area/volume of the damaged zone calculated from both simulation and in-vitro experiment fits well and the deviation is small. One of the main reasons for the deviation is the lack of accurate values of the tissue optical properties. In further experiments this needs to be validated.	Introduction Laser radiation is now used routinely in surgery to incise, coagulate, or vaporize tissues. The laser light power is converted into heat in the target volume with ensuing coagulative necrosis, secondary degeneration and atrophy, and tumour shrinkage with minimal damage to surrounding structures [ 1 ]. The use of lasers in surgery introduces some desirable features over normal surgical methods such as increased precision, improved hemostasis, and less tissue manipulation. Laser light power is thereby delivered to the targeted area by an optical fibre. The use of an optical fibre as applicator for interstitial light delivery was demonstrated, among others, by Bown [ 2 ] in 1983 as a method of heating and destroying deep-seated tumours [ 3 ]. The biological effects of laser energy depend on the laser wavelength, laser power, the duration of irradiance, blood perfusion, and both the optical and thermal properties of the tissue involved. Laser-tissue interaction mechanisms may be thermal, photochemical, or mechanical in nature [ 4 ]. Photochemical like PhotoDynamic Therapy (PDT). Mechanical like the effects induced using pulsed lasers (photoacoustic, photodisruptive). The surgical procedures that involve coagulation or ablation of tissue are thermal. Clinical studies have yet to demonstrate that LITT is practical for the palliation of hepatic and nasopharyngeal tumours (e.g. [ 5 - 9 ]). The criteria for the clinical success of the thermotherapy for tumours in homogeneous tissues, for example, in the liver or brain, was described, among others, by Vogl [ 3 ], who placed applicator/diffuser at the centre of the tumour, using MRI online monitoring of the thermal changes to control the treatment [ 6 ]. In contrast to homogeneous and simple tissues, however, normal anatomical structures are more complicated. Especially for small tumours near main vessels a positioning of the laser applicator at the centre of the lesion is difficult and maybe not the optimum choice. Modelling the laser-tissue interaction is beneficial for the analysis and optimisation of the parameters governing planned laser surgical procedures. Nevertheless, we still lack an adequate model that grants accuracy. Most of the models suggested depend greatly on simplifications of the real problem, either in the geometry they offer or in the system of equations they use. Some models, which use the bioheat equation, neglect the role of the changes in the tissue properties during temperature elevation process [ 10 ], which deem such a model unrealistic, especially considering high temperatures. Few modelling methods have simulated the behaviour of LITT in human tissue. Best known is the Monte-Carlo method described, among others, by Roggan et al. [ 11 ]. This method can simulate the use of multiple applicators but is limited to symmetric geometries and has not been correlated to real anatomic datasets. Similar to this is the finite difference method described in [ 12 ], where the authors did not include the coagulation process with its irreversible changes in optothermal tissue properties. In order to overcome these limitations, we included the damage function as well as perfusion terms in the modelling process, taking dependencies of these parameters into account. This paper describes in detail the bases for a modelling method to simulate the effect of LITT for the treatment in various indications near large vessels, such as the carotid artery in the neck region. We thereby propose the use of LITT, frequently applied in the treatment of liver tumours [ 6 ], in more sensitive areas such as the neck region to treat tumours in lymph node metastases or epithelial carcinomas with dimensions of 0.5 cm – 2 cm in diameter. The actual response of tissue to laser irradiation is a time-dependent phenomenon. Initially, there are thermal and possibly photochemical changes of the tissue at the molecular level. Next are changes in tissue perfusion caused by thermally induced vascular relaxation and/or vessel damage. Heat deposited at the application site is transferred to adjacent structures. This may be desirable for coagulation purposes – or it may cause unexpected thermal damage to otherwise viable tissues adjacent to the irradiation site. The rate of heat transfer depends on the composition and organization of tissues involved. Blood perfusion during and after irradiation has significant effects on the size of the damage zone. We discuss in this paper our mathematical approach, its considerations and restrictions. In the main part we present the mathematical and physical backgrounds used to achieve the model. Then we present and discuss the results of our simulation in comparison with the results of our in-vitro experiment. Materials and methods Our model of LITT considers both optical and thermal effects. It is based on calculations describing the light distribution using the diffusion approximation of the transport theory; the temperature rise using the bioheat equation, including the effect of microperfusion in tissue to determine the extent of thermal damage, and the dependence of thermal and optical properties on the temperature and the injury. Injury is estimated using a damage integral, which depends on the temperature elevation history. The order and flow of the modelling steps are described in the following sections in detail. The geometry of the 3D model The head and neck area consists of complex anatomical structures in close proximity. In sonographic 3D volume datasets of the neck area the sternocleidomastoideus muscle and the neck vessels (common carotid artery, internal carotid artery, external carotid artery and the internal jugular vein) serve as leading structures [ 13 ]. Because of the almost superficial anatomic position of the vessels and their straight course (Fig. 1a ) especially the common carotid artery is easily shown sonographically. Differentiation of inflammed lymph nodes and metastases located parallel to the large neck vessels [ 14 ] can be achieved 90% of the time with the help of signal-enhanced colour duplex sonography [ 15 , 16 ], making ultrasound preferable to other imaging techniques. Figure 1 The geometry used. The left carotid artery is shown in this figure labelled with ca . The following letters indicate the orientation: s for superior, i for inferior, p for posterior, a for anterior, l for left, r for right. (a) is a photo of the human anatomy in the neck area. The carotid artery is shown here after moving the vein to the cranial direction. (b) shows the corresponding freehand 3D ultrasound dataset of the human neck region acquired axially. The 3D image in the top of (b) shows the 3D ultrasound volume together with the carotid artery segmented with 3D Slicer software [17]. The 3D model is displayed with 40% transparency. (c) displays the model used in the simulation approximated according to the geometry of the human neck shown in (a) and (b). The segmentation of the 3D ultrasound dataset in (b), is available as a video stream, too, showing the geometry of the carotid artery (see Additional file 1 ). One can easily segment the carotid artery from 3D sonographical, MRI or CT volume datasets. For the segmentation of our 3D ultrasound dataset (Fig 1b ) we used the software package 3D Slicer [ 17 ]. The real geometry is complex and needs to be simplified to reflect limited computational capacity. We thus obtained the 3D base model consisting of a cube of 447 cm 3 shown in Fig. 1c . The cube contains the blood vessel approximated either by the cylindrical shape or much better by a shape of cone. The applicator is shown as a thin tube of 2.5 cm perpendicular to the vessel (Fig. 1c ). Commercially available laser applicator fibres for thermotherapy frequently have a water jacket to cool the surface. The applicator is assumed to be a cylinder, and the cooling effect is implemented as a boundary condition at the diffuser surface. The tissue surrounding the vessel is treated as a homogeneous muscle tissue. According to the geometry described using a mesh is generated to perform a finite element method calculation (Fig. 2 ). The model was implemented with FEMLAB 2.3 as an add-on to Matlab 6.5 for finite element modelling [ 18 ]. Figure 2 The finite element method (FEM) mesh. The values of each variable and of each property value are evaluated at each point of the mesh. In other words, the "simulation loop" in Fig. 3 is performed at each of point of the mesh. The bi-points variables' values are evaluated using an interpolation process. The light distribution equation In most tissues, both absorption and scattering are present simultaneously. A mathematical description of the absorption and scattering characteristics of light can be performed analytically or by using the transport theory. Transport theory has been extensively used when dealing with laser-tissue interactions. Furthermore, experimental results have confirmed its validity in most cases [ 19 ]. The photon propagation described using the transport theory has been dealt with already in [ 4 , 19 - 23 ]. Exact analytical solutions to the radiative transport equation have been found for only few special simple cases. However, when scattering processes dominate absorption in the medium, a high penetration depth of the light is the consequence. This is the case in LITT treatment in human tissue using an Nd:YAG 1064 nm laser or even a diode 830 nm laser: The penetration depth of both types ranges from between 1300 μm – and 1400 μm, whereas the penetration depth of other laser types like argon 514 nm laser or CO 2 10600 nm laser is less than 350 μm. This leads us to the possibility of applying light diffusion approximation to the transport theory [ 4 ]. Because of the high penetration depth of the Nd:YAG laser in turbid media, diffusion theory provides a relatively accurate description of light propagation. In three dimensions the diffusion equation needs to be solved numerically, because an analytical solution is not possible [ 4 ]. FEM is the most practical method; moreover a number of different efficient solutions using FEM are now available. An exact derivation of the light diffusion equation can be found in [ 4 , 20 ]. Here, we give the light diffusion approximation to the transport theory, which is implemented in our model (Fig. 3 : Light Distribution Equation): Figure 3 The simulation loop. The figure shows the simulation flow chart as a step in the imagined surgical planning system (left side). The future goal of the surgical planning system is to verify the three parameters governing a laser treatment: the applicator position, the laser power, and the exposure duration. The temperature starting point of the volume is set to normal body temperature. Three input parameters are taken: average blood velocity, laser power, and application time (right side: input). The main part of the simulation is the loop, which calculates the variables in the forward steps, then updates the values of the different properties (parameters) in the backward step according to the results of the forward step. The loop follows the section materials and methods and uses its nomenclature for variables and functions (right side: loop). The output of the solver consists of three parts: the light energy fluence rate φ ( r . t ), the temperature distribution T ( r , t ), and the damage Ω( r , t ) (right side results). The results are explicity shown in figures Fig. 4 and Fig. 5. The roman numbers (I-VI) refer to the equations in the text. φ being the light fluence rate [W cm -2 ], D the diffusion coefficient [cm], and Q the source term [W cm -3 ]. μ a is the absorption coefficient and μ ' s the reduced scattering coefficient in tissue. The roman number (I) indicates the position in Fig. 3 . The relationship between the reduced scattering coefficient and the scattering coefficient, μ s is described by μ ' s = μ s (1-g) , with g being the anisotropy factor incorporating the effects of directionally dependent scattering. The absorption coefficient μ a for visible and for near infrared radiation ranges between 0.001 mm -1 < μ a < 10 mm -1 for biological tissues. While for the scattering coefficient μ s is in the order of 1 mm -1 < μ s < 100 mm -1 [ 20 ]. The optical properties μ a and μ ' s depend on the tissue, and they change their values during a real treatment. In order to simulate this effect, the optical properties of the tissue are functions of the damage Ω [ 20 ] (see section on "damage function"). The damage function Ω describes the pathologic state of the tissue during treatment. In general, for simple geometries like a point-source, the solution of the light diffusion equation will be an exponentially decreasing function with effective attenuation coefficient given by: As an example, the solution of eq. 1 for a light source similar to a medical applicator in a homogeneous medium takes the shape of an ellipsoid [ 24 ] as shown in Fig. 4 . Figure 4 Solution of the light distribution equation. Left: the energy density colour index. The solution is elliptical as expected. The penetration depth in the vessel is less than in the tissue, because of different scattering and absorption coefficients. The heat distribution equation in tissue The aim of irradiation with laser energy is to produce heat in the targeted tissue. Excess heat is either stored or extracted, leading to changes in the local temperature. The bioheat equation was repeatedly used to describe the heat changes in biological tissue [ 4 , 19 , 20 ]. The bioheat equation is the realizing of the principle of conservation of the energy applied to tissue volume, (Fig. 3 : Heat Distribution Equation, in Normal Tissue), where T is the temperature [°C], ρ the density of tissue [kg cm -3 ], c the specific heat of tissue [J kg -1 °C -1 ], k the thermal conductivity of tissue [W cm -1 C -1 ], r the position vector [cm], t the time [s], T art the temperature of arterial blood [°C], S the deposited light power [W cm -3 ], w b [ml/(g.min)] is the tissue average volumetric blood perfusion rate (but because the density of blood ρ b is to be considered as a constant value, it is possible to call ρ b · w b [kg s -1 cm -3 ] the average volumetric blood perfusion rate , unfortunately usually denoted in the literature also with w b ), and c b the specific heat of blood [J kg -1 C -1 ]. The coefficients ρ , k and w b are functions of temperature T . As a basis for the optical and thermal parameters for the simulations, we used values published by Mueller et al. [ 1 ]. Especially for normal body muscle tissue the physical properties are collected in Table 1 . Table 1 Listing of physical parameters. The table shows the physical parameters for native tissue and for coagulated-tissue, as well as for blood [25]. The small letters indicate the references where the material parameters are taken from: a) [20], b) [26], c) [27], d) [28], e) [29]. Muscle Blood Native Coagulated Absorption coefficient, μ a (cm -1 ) a) 0.23 a) 0.22 b) 0.44 Scattering coefficient, μ s ' (cm -1 ) a) 1.3 a) 13 b) 2.78 Density, ρ (kg cm -3 ) c), d) 1.04·10 -3 d) 1.06·10 -3 Specific heat capacity, c (J kg -1 °C -1 ) d) 3.64·10 3 d) 3.89·10 3 Heat conductivity, k (W cm -1 °C -1 ) c) 5.18·10 -3 e) 5.4·10 -3 The heat distribution equation in large vessels 1. The incompressible Navier-Stokes equation In order to make the model adaptable to individual shapes of segmented vessels, we considered the geometry of a large vessel as a volume in which an incompressible fluid (blood) flows. The direction of the blood flow and the initial speed profile are implemented as boundary conditions. The incompressible Navier-Stokes equation for the blood (Newtonian fluid) reads: Here, η is the dynamic viscosity [kg s m -1 ], ρ the density [kg m -3 ], u the velocity field, p the pressure [N/m 2 ], and F a volume force field such as gravity. Implementing the Navier-Stokes equation in the model allows us to present a time-periodic change in the blood flow rate, i.e., to simulate the beat cycle effect in the vessel. The main effect here on the result of the simulation lies in the accuracy of the estimated heat elevation in the tissue: A continuous blood flow has a different profile than the cycled flow (laminar and not laminar, or rather complex with four beat phases), which yields a different final cooling effect. For vessels away from the heart, the pumping cycle does not clearly appear; it tends to be a normal laminar flow. In this case ( u ( r , t ) = u ( r )) and eq. 5 is reduced to the following: 2. The bioheat equation in large vessels The heat convection between tissue and a large vessel occurs as a direct energy transfer rather than perfusion. The vessel is a heat sink in the treated volume. Therefore, the perfusion term in the bioheat equation has to be modified to consider heat conduction and blood flow. A new term, the so-called enthalpy transport , is added to retain the validity of the bioheat equation. The term serves for interpreting the internal energy flow out of the control tissue volume by means of the blood flow [ 30 ]. Considering the blood velocity field ( ) in a large vessel the bioheat equation becomes: The damage function The thermal damage in cells and tissue is described mathematically by a first-order thermal-chemical rate equation, in which temperature history determines damage. Damage is considered to be a unimolecular process, whereby native molecules transform into a denatured/coagulated state through an activated state leading to cell death [ 4 , 19 ]. Damage is quantified using a single parameter, Ω, which ranges on the entire positive real axis and is calculated from the Arrhenius law: where A [s -1 ] is the frequency factor, Ea [J/mole] the activation energy, R [J mole -1 K -1 ] the universal gas constant, and T [K] the temperature. C( r , 0) and C( r , τ ) are the concentrations of the undamaged molecules at the beginning and at time τ , respectively. Damage Ω (eq. 8, Fig. 3 , Damage Function) is dimensionless, exponentially dependent on temperature, and linearly dependant on time of exposure. The activation energy E a and the frequency factor A are derived from thermodynamic variables. They describe the denaturation process of proteins and other cellular constituents. A ranges from 10 40 s -1 to 10 105 s -1 , and E a from 10 5 J/mole to 10 6 J/mole [ 4 ]. The equation above indicates that the measure of damage describes the probability for tissue being destructed. It is the logarithm of the ratio of the initial concentration of undamaged tissue to the concentration after damage has accumulated, for the time interval t = 0 to t = τ . Therefore, Ω = 1 corresponds to the reduction in concentration of native molecules to a 37% level for a unimolecular system – an irreversible damage of 100% of the affected cells. However, in terms of the thermal damage to tissue, Ω ( r , t) is a function of the observer's definition of damage. In [ 31 ] a limit of Ω >0.6 has been discussed as a margin of final tissue destruction (Fig. 5 ). A value of Ω = 0.6 corresponds to reduction in concentration of native molecules to 50% level. Figure 5 The heat distribution and the damage in the volume. Left: the heat colour index in °C. The damage appears when the value of the damage function Ω (eq. 8) reaches the threshold of 0.6. Here the image shows the results after 200 s. The damage zone is shown in grey. Fig. 5 is available also as a video stream demonstrates the temperature rise inside the tissue. The video stream shows where, how, and when this damage appears (see additional file 1 ). The dependences of the tissue properties on tissue temperature and damage Heat capacity is assumed to be constant over a wide temperature range. The temperature dependence of thermal conductivity and density is taken into consideration by the following linear approximations [ 20 ] (Fig. 3 : Properties): On the other hand, the optical properties are influenced directly by the damage Ω. The scattering coefficient of coagulated tissue is much higher than that of native tissue. The optical properties change during the process of tissue coagulation, leading to a higher scattering coefficient and a nearly constant absorption coefficient. This becomes obvious by the change of the colour of the irradiated area (bleaching) and leads to reduction in penetration depth. The actual property set is calculated from the actual damage value as well as the optical properties in the native and coagulated tissue states [ 20 ] (Fig. 3 , Properties): Here, μ s , native and μ s , coagulated denote the scattering coefficients of native and coagulated tissue, respectively, g being the anisotropy factor. In literature g is mostly considered constant. Anyhow, some authors [ 20 , 31 ] reported the possibility of different values for g according to the damage state. Model implementation The diagram in Fig. 3 illustrates the flow of the simulation. There are three main parts to the modelling of laser-tissue interaction (Fig. 3 , right part): • First, the irradiance distribution in the different tissues is determined by directly applying eq. 1. As shown in equations eq. 11, eq. 12, and eq. 13 the values of the properties depend on damage Ω (eq. 8). In the first loop step Ω is zero, and it starts to increase according to the rise in temperature, i.e., the different optical properties have as their starting point native tissue and as end point coagulated tissue. The actual value lies between both limits as determined according to Ω. • In the second step, the temperature distribution in the tissue caused by laser energy deposition is estimated by solving the two bioheat equations for tissue and large vessel. The source term in both equations is defined by the absorbed energy at each mesh point (Fig. 2 ) from the distributed light energy calculated in the first step (light-energy to heat-source coupling) using: • A by-step here is the estimation of the blood speed field ( ) from the Navier-Stokes equation (either eq.5 or eq. 6). In our solved model, because we suggested treatment as taking place in the neck near the carotid vessel, we considered to use eq. 6 for obtaining the speed field, which is valid for laminar flow. • In the third and main part, thermal damage is predicted from spatial and temporal temperature distribution (eq. 8). • After estimating the heat distribution and the damage value, we perform a backward step to calculate the new values of the properties according to eq. 9 through eq. 12, which are updated in the equations set for the next loop iteration. For our calculations we used FEMLAB's standard mesh generator with its default settings for modelling [ 18 ]. The mesh consists of 5354 nodes, more dense near the applicator and becomes coarser moving towards the walls. The time-dependent equation set described in the previous sections was solved using FEMLAB's time-dependent solver "femtime". The default settings for the solver were used: 0.01 for relative error tolerance and 0.001 for absolute error tolerance. Because of the non-linearity of this problem, the special time stepping algorithm "fldaspk" offered by FEMLAB was used in order to obtain a stable and convergent numerical solution [ 18 ]. A normal numerical solution to initial value problem of differential equation generates a sequence of values for the independent variable (time) t n and a corresponding sequence of values for the dependent variable ( φ , T , u , Ω, and all other variables depend on them in this case) so that each φ n , T n ,... approximates the solution at t n [ 32 ]. Modern numerical methods automatically determine the step sizes h n = t n +1 - t n so that the estimated error in the numerical solution is controlled by a specified tolerance [ 32 ]. The "fldaspk" solver uses the algorithm of the known DASPK solver written by Linda Petzold, which uses variable-order variable-stepsize backward differentiation formulas (for independent variable, time t in this case) [ 33 ]. There is no control on the time step itself, rather on the specified tolerance of each variable. Light is considered to be emitted from an interstitial fibre with a fibre-diameter of 1 mm; it was modelled as an isotropically radiating cylindrical source (Fig. 4 ). In the real treatment a cooling process using a special cooling catheter is performed to keep the temperature at the surface of the applicator low, preventing damage at its surface. A special boundary condition at the applicator surface should be applied in order to simulate this cooling effect. In the model this is realized by setting the outer surface temperature of the applicator to a constant value (normal body temperature T = 37°C). At the modelled volume surfaces the insulation boundary conditions, optical and thermal, are used, where n is the outward unit vector normal to the surface. This means the gradients of light fluence rate and of temperature vanish at the surface. Even though this condition is more suitable for light fluence rate, as small amount of radiations reach the surface, but in general the temperature and light fluence rate will not be constant. The need to set this condition in this way is because that the numerical solver demands defined fixed boundary conditions, which sometimes do not agree with the real situation. According to the NCRP-data [ 34 ] the perfusion rate w b over the entie tissue is set to 1.4·10 -6 kg of blood s -1 cm -3 for T < 60°C and to 0 when T ≥ 60°C, which is to be considered as a normal result of stopping the blood perfusion according to temperature elevation in the tissue [ 31 ]. In order to solve the Navier-Stokes equation, we set the dynamic viscosity, η , to 3.5·10 3 kg·s·m -1 . To evaluate the damage, Ω, the activation energy E a is set to 670000 J/mol and the frequency factor A to 9.4·10 104 s -1 [ 22 , 35 ]. Damage is considered to appear when Ω reaches the value of 0.6 [ 31 ]. The simulation takes around 2 hours on Sun Blade 2000 with Solaris 9 OS, 6 GB RAM and Ultra SPARC IIIi processor. Experimental validation We performed a single in-vitro experiment to check our model. The setup is shown in Fig. 6 . The experimental work was performed using a fresh piece of porcine muscle tissue. In order to simulate the cooling effect of the blood flow in a vessel, a transparent tube (Polyethylene) and a porcine blood were used. An electronically controlled roller pump system (Storz Endomat LC203303) was used to pump the blood through the tube. We used the online ultrasound sonography to situate the applicator in its right position in front of the tube and to get the 3D ultrasound data analogy to the base geometry shown in Fig. 1 . The sample has dimensions of 1265 cm 3 . The tube inner diameter is 5 mm and the outer diameter is 7 mm. The blood and the laser cooling liquid had the room temperature of 21.4°C while the sample itself 17.6°C. A laser power of 30 W and blood average flow speed of 40 ml·s -1 were used. We measured the exact distance between the laser applicator and the tube edge, 3 mm, at the end of the experiment after performing the cut in the sample. We fixed our application time to 300 s. Figure 6 The experiment setup. 1- laser applicator, 2- transparent tube (considered as a blood vessel), and 3- is ultrasound sonography prob. The tube was introduces by pulling it through a hole made using dipped knife. The online sonography was used to verify the position of the applicator. In the simulation model we omit the perfusion term, as there is no perfusion to be considered in-vitro. We simulated the tube (blood vessel) with diameter of 6 mm. All other experimental conditions are implemented in the model as they are in the experiment. Because of the lack of data, which describe the properties of the porcine tissues, in all literature available to us, we used the same data presented in Table 1 to complete this simulation. The highest temperature and widest damage are reached in front of the centre of the applicator. Taking into account the perpendicular surface to the applicator at centre, which is the most critical in the volume as all effects participate together: applicator, vessel, and blood flow, we complete the comparisons of the results between the experiment and the simulation using this plane. Hence, we made a cut in the probe at the level equivalent to this plane. Coagulation of tissue is immediately apparent and always indicates lethal thermal effect [ 4 ]. Anyhow, and in order to comment on, a damage boundaries have to be identified. For most tissues, coagulation can be seen with naked eye as whitening of the tissue associated with turgor and opacity [ 4 ]. The damage boundary in the probe cut is determined by applying a grey threshold of 50% on the picture of the cut after changing it to grey scaled image. Away from the applicator position in a certainly undamaged area, applying this threshold led to 92.68% of the pixels to be black and 7.32% white, while in the damaged tissue 6.64% of the pixels were black and 93.3% white. The Matlab 6.5 R13, its Image Processing 4.0 Toolbox, and Paint Shop pro 8 were used to perform these steps. A comparison in the z-direction needs an up-down cut (y-z plane) through the applicator position perpendicular to the tube/vessel, which was not possible after the cut in x-y direction. Results In [ 25 ] we presented results from different geometries. Here, we focus on the results from the physical and mathematical points of view. We present also a comparison between the results of our model and the results of the in-vitro experiment. Theoretical results Fig. 4 shows that the light energy is distributed, as expected, in a shape of ellipsoid, because of the absorption in the tissue. It also shows that the light penetration depth in blood is less than in the normal tissue, because of the different values of the absorption and scattering coefficients. Fig. 5 shows the heat elevation. The cooling effect of the blood vessel may be clearly identified here. Fig. 5 is available as a video stream, too. This video stream demonstrates the temperature rise inside the tissue as a result of the irradiation. The video stream shows where, how, and when this damage appears (see Additional file 2 ). The dimensions of the damage zone, which may be considered the target goal of the simulation, can be calculated directly by producing grided axes in all the 3D and 2D results as well as with routines written especially for this aim. Comparison with experimental results Fig. 7 shows the interpreted development of the damage zone. The calculation shows that the damage starts with an application time of 72 s using laser power of 30 W. It starts at a distance of 2.5 mm from the centre of the applicator in the opposite direction of the vessel. After that, the damage zone will spread to all directions as it is shown in Fig. 7 . Fig. 8 shows the solution of the heat and light distribution equations for this model. In the first 60 s there is no noticeable change in the light distribution. In the next 30 s the damage zone appears, which leads to changes in the scattering and absorption coefficients (eq. 11 and eq. 12). Figure 7 The interpreted development of the damage zone as a result of the simulation model using the experimental conditions. The applicator is recognized as a grey point in the middle and is indicated with (App). The damage zone starts with an application time of 72 s. The damage zone is presented here at 73 s, 120 s, 180 s, 240 s, and 300 s of application time as indicated in the figure. Figure 8 Dispersion of heat and light distribution considering the experiment conditions. The figure shows the results at 60 s, 90 s, and 300 s of application time. The isolines show the light distribution, which represent the deposition of the energy irradiated from the laser applicator. The major change occurred to the energy penetration depth happens between 60 s and 90 s of application time. The dashed area presents the damage zone at these times. Fig. 9 shows the overlaid damage zones of both model and experiment. The kidney-shaped lesion is about 21.2 cm 2 , while our model shows a damaged zone of 2.11.45 cm 2 . We obtain calculation errors of 5% in the x-axis parallel to the blood vessel, and of 20% in the y-direction perpendicular to the vessel. This deviation happens mainly due to inaccurate optical properties values. Figure 9 The results from simulation and experiment. For the experiment the following settings were used: time of application: 300 s, laser power: 30 W, blood flow rate: 40 ml/s, and applicator-vessel edge distance: 3 mm. Using these conditions a simulation is performed. In the figure the damage zones of both model and experiment are overlaid to show the deviation. The black oval shows the damage zone obtained from the simulation using the above conditions. The two parallel black lines indicate the vessel position and diameter in the simulation. The lesion is about 21.2 cm 2 , while the calculated damage zone is 2.11.45 cm 2 . The calculation error ranges from 5% in the x-axis parallel to the blood vessel to 20% in the y-direction perpendicular to the vessel. Both cutting the tissue with scalpel and the opening induced a tissue movement. This movement is a reason for the error in y-direction beside the error obtained from the optical properties, which affect all directions. The discoloration in the top of the image at the opposite side of the artificial vessel is due to the cut and the opening of the probe. Discussion To date most simulation models of LITT have used the Monte-Carlo method (MC) to calculate the light distribution, then combine its results with Finite Difference method (FDM) to calculate the heat distribution. Because of its formulation, this combination fits very well for a radially symmetric problem. A weakness arises, however, when dealing with asymmetric volumes in real human anatomy. Arbitrarily curved surfaces separate the different tissues. Consequently, calculations from the FDM becomes so complex that errors start to appear in the results presented stemming from the dependency of FDM on dividing the volume into voxels. One way to overcome this is to increase the voxels' number. Indeed, this leads to less error at the tissue-separating surfaces, although it increases the resources and calculation steps, making the procedure inconvenient. In principle, combining MC and FEM (instead of FDM) is possible theoretically, and seems to be promising as it overcomes the latter problems, but to our knowledge has not implemented yet. From another perspective, MC solution converges to the exact solution of the transport equation only when the number of traced photons increases infinitely [ 4 , 21 ], which yield time consuming calculations. Because we are dealing with an asymmetric geometry, we chose the FEM. It allows us to define and refine the mesh in the volume of interest in order to obtain more precise results. Furthermore, using a FEM mesh we are able to adapt individually the mesh for each patient's dataset. Never the less, it was not necessary to combine methods, FEM is used for all equations. The model we propose depends on the following considerations: • The coupling of a set of time-dependent equations, which simulate the whole process of the LITT treatment (Fig. 3 ). The set of equations includes the light diffusion equation, the bioheat equation, the Navier-Stocks equation, the damage function, and the dependencies of the properties on temperature and ensued damage. • We consider the functional dependence of the various tissue properties at the various spatial and temporal points, according either to the tissue type, or temperature, or the damage value – or even a combination thereof. • We take into account the irreversible changes in the tissue stemming from the treatment as they directly affect the solution of the set of equations. Our model remains a mathematical model, meaning errors could appear from the considerations and simplifications made to realize it. Generally, such errors appear because of the following reasons: • The inaccuracy of the optical, thermal, and damage properties are main point in the model's set of equations. In fact, these properties play a key role in the accuracy of the model's results. Many methods have been presented to calculate these properties [ 4 , 19 , 21 ], but still we see differences in the values presented by the different groups, which reflect the difficulty of measuring these properties. The problem is increased by the dependencies of the properties on the different variables (temperature, damage) over time. This makes the deviation neither linear nor regular. • An error appears because of machine performance limitations: The available memory limits the number of mesh nodes and the degrees of freedom (DOF) used to build the model. This causes a deviation from an otherwise accurate result [ 18 ]. On the other hand, it is useless to increase the nodes number or the DOF arbitrarily: This would result in more time consuming calculations, since one would always have to run an interpolation and smoothing process as next to last step. In practice, a suitable number of nodes/DOF should be chosen, so that the interpolation routines estimate smoothly the value of all variables between nodes. Our model is based on what was mentioned above, with FEMLAB's standard refining processes at the critical areas (around the diffuser and vessel). It is important to refine the mesh around such surfaces, something that can be done much more conveniently using FEM rather than FDM. The convenience of the FEM-based modelling may be found in this very point of its ability to have different degrees of refining in the mesh according to how critical the region is. • Absolute tolerance: All numerical methods have an allowed error (absolute tolerance) that reflects the criterion of the convergence. Normally, different solvers use different tolerances. In our model we used the FEMLAB's default tolerance value of 0.01 which leads to a final error of 1%, considered as a reasonable value for modelling. One way to follow these errors and deviations from a real treatment result is to estimate them and to eliminate their effects from the final results of the model. This can be realized and implemented in the model by adding an error-correcting factor from the first degree (or even higher) in the set of equations correcting the result of each equation at each time step. These corrective factors should be measured practically by comparing the results of the model and the results from real experiments on test tissues or probes. Our experiment shows a deviation of 5% in x-direction and 20% in y-direction. As the main reason for this deviation we propose inaccurate values for the optical tissue properties. Fig. 7 and Fig. 8 clearly show the kidney-shaped damage zone caused by the cooling effect of the blood vessel (or the tube in the experiment), which keeps its neighbouring in the native state for longer time. In literature [ 20 , 31 ], the value of the absorption coefficient μ a for both native and coagulated different biological tissues are close. Baring that in mind, and knowing that the value of the scattering coefficient μ s becomes normally, for biological tissue, 10 times greater than its starting value, i.e. native state, we can judge that as soon as the damage zone appears and the moving from native to coagulated state according to eq. 11, eq. 12, and eq. 13 the deviation in the calculations will increase as well. Thus, accurate values of the different tissue properties, and especially the optical properties are key points in obtaining realistic results from the simulation. One promising technique for determination of optical properties was presented by Dam et al. [ 36 ]. There method provides an online values of the optical properties at 660 nm, 785 nm, 805 nm, 974 nm using a cylindrical probe head situated on the skin. It still require further researches. Anyhow, thinking of developing such a method to be able to gather information about optical properties at 1064 nm interstitially using a catheter during the irradiation can be of great value for modelling. Our model gives the possibility to implement such a gathered information directly. Thinking of using it online to predict the damage and controlling the irradiation power needs for sure more researches. Finally, beside the error obtained from the optical properties, which affect all directions, both cutting the tissue with scalpel and the opening induced a tissue movement. This movement is a reason for deviation, especially in y-direction, as we perform the cutting in this direction. Conclusions For several years now LITT has been a well-known and approved therapy system for tumour ablation in the liver and some other anatomical regions. Minimally invasive LITT procedures use a Nd:YAG 1064 nm laser. Therapy planning, however, remains unsolved and is still a challenging issue. Today's simulations are based on symmetric geometries. Without exact therapy planning systems, the usage of LITT is limited to homogeneous tissues or the respective surgeon's experience. The finite element technique proposed in this paper can overcome both limitations. We propose a model to validate in the future the LITT method in other anatomic regions. The model enables the efficient simulation for predicting the damaged zone induced with the diffuser of the LITT. The simulation is performed for tissue ablation near vessels, though obviously FEM is not limited to this. Exemplarily, we implemented the model for tissue ablation near the carotid artery in the neck region using an approximation for the artery shape. We describe the bases necessary to calculate the effects of the temperature rise caused by the absorption of light energy in the tissue, using the bioheat equation and including the cooling effects of vessel blood flow and micro-perfusion in tissue in order to determine the extent of thermal damage. The shape of the carotid artery is derived from a real segmented geometry based on, but not limited to, 3D ultrasound. Experimentally, we performed a laser irradiation in a porcine muscle tissue sample. The results of our model diverge between 5% to 20% from the lesion obtained in the experimental work. From the authors' point of view two major reasons can be identified. The lack of accurate data describing the thermal and optical properties leads definitely to deviations. Furthermore the cut of the probe with scalpel induces a certain tissue shift, especially in the y-direction. Anyhow, more experiments with different conditions are necessary to be able to carry out a statistical study and find the exact origin of the deviation, and, if necessary, define an error correction factors and add them to equation set. But that does not set aside the desire of accurate values for the properties of the tissue. From another hand, still our model practical, it presents a step in using segmented data as basis for much more detailed surgical therapy planning. Combining LITT and adequate planning system could increase both the anatomical application range and the quality of therapy procedures. Supplementary Material Additional File 1 Animated gif file, the Geometry . The animated gif shows the 3D ultrasound volume together with the carotid artery segmented using 3D Slicer software [ 17 ]. The movie belongs to Fig. 1b . The gif file can be played using the internet browser. Click here for file Additional File 2 Animated gif file, The heat distribution and the damage zone in the volume . The video stream demonstrates the temperature rise inside the tissue. The video stream shows where, how, and when this damage appears. The damage zone is shown in grey colour. The gif file can be played using the internet browser. Click here for file	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC546235.xml
521084	Anogenital distance in human male and female newborns: a descriptive, cross-sectional study	Background In animal studies of the effects of hormonally active agents, measurement of anogenital distance (AGD) is now routine, and serves as a bioassay of fetal androgen action. Although measurement of AGD in humans has been discussed in the literature, to our knowledge it has been measured formally in only two descriptive studies of females. Because AGD has been an easy-to-measure, sensitive outcome in animals studies, we developed and implemented an anthropometric protocol for measurement of AGD in human males as well as females. Methods We first evaluated the reliability of the AGD measures in 20 subjects. Then measurements were taken on an additional 87 newborns (42 females, 45 males). All subjects were from Morelos, Mexico. Results The reliability (Pearson r) of the AGD measure was, for females 0.50, and for males, 0.64. The between-subject variation in AGD, however, was much greater than the variation due to measurement error. The AGD measure was about two-fold greater in males (mean, 22 mm) than in females (mean, 11 mm), and there was little overlap in the distributions for males and females. Conclusion The sexual dimorphism of AGD in humans comprises prima facie evidence that this outcome may respond to in utero exposure to hormonally active agents.	Background In animal studies of the effects of hormonally active agents, measurement of anogenital distance (AGD) is now routine [ 1 - 16 ], and serves as a bioassay of fetal androgen action. In rodents, perineal growth is dihydrotestosterone-dependent [ 17 ], males have a greater AGD than females, and use of AGD to sex newborns is standard [ 18 ]. In animals AGD is correlated at only modest levels with body weight [ 19 ], because these measures reflect the effects of endocrine axes that are largely independent. AGD usually tracks through life, varies by dose of antiandrogen, and can be predictive of other androgen-responsive outcomes [ 20 ]. Although measurement of AGD in humans has been discussed in the literature [ 19 , 21 - 23 ], to our knowledge it has been measured formally in only two descriptive studies of females [ 24 , 25 ]. Because AGD has been an easy-to-measure, sensitive outcome in animal studies, we developed and implemented an anthropometric protocol for measurement of AGD in human males as well as females. This work constitutes a modest step towards evaluation of AGD in human males as a potentially useful anthropometric measure and indicator of in utero androgen status. Methods Subjects A cross-sectional study was conducted among the newborn children of women admitted for delivery to the Dr. Ernesto Meana San Román General Hospital in Jojutla, Morelos, Mexico, in 1999. This hospital provides medical care to low socioeconomic status and uninsured populations. The study included 87 newborn infants, none of whom had congenital defects or had been admitted to the neonatal intensive care unit. All infants were born at term (≥38 weeks gestation), except for one (32 weeks). The infants were of both sexes and were born after spontaneous cephalic delivery or caesarean section. Within 6 hours of birth, a structured questionnaire about family background and obstetric history was administered to the mothers, and anthropometric measurements were taken on the newborns. Anthropometry Anthropometric measurements were taken of weight, length, head circumference, and AGD. AGD was measured as follows: the newborn infant was in the dorsal decubitus position; both hips were flexed and light pressure was exerted on the infant's thighs until the examiner's hand touched the subject's abdomen. Measurements were made with Vernier calipers. Distance was measured from the center of the anus to the posterior convergence of the fourchette (where the vestibule begins) in female infants [ 24 ]; and from the center of the anus to the junction of the smooth perineal skin with the rugated skin of the scrotum in male infants (Figure 1 ). Gestational age was estimated according to the Dubowitz scoring system [ 26 ]. Figure 1 Schematic Diagram of Measurements Done, by Sex Reliability Before any contact with the 87 subjects in the main study, the personnel performing the anthropometry examined 20 other neonates; all of whom were born after ≥38 weeks gestation. In this standardization training, 7 female infants and 13 male infants were measured twice by each observer. A sufficient time interval (30 minutes) was allotted between each measurement so that the second would not be influenced by the observer's memory of the first. These data were used to examine the reliability of measures and sources of variance. Statistical analysis The reliability of the anthropometric measures was calculated as the Pearson correlation coefficient between the paired measures. The observations taken by the two observers were not statistically different when compared using a paired t-test (results not shown). Analysis of variance (ANOVA) with a random effect term for subject was used to estimate between-subject, between-observer, and within-observer components of variance, by sex. For the main study, a linear regression analysis was used to evaluate birth weight, birth length, and gestational age as predictors of AGD. Age of the mother, number of pregnancies, and time elapsed between birth and measurement were not important predictors (or confounders) of AGD in univariate or multivariate models and were not considered further in the analysis. To examine influential values and the overall fit of the model, we conducted an analysis of residuals, but found nothing of note. The protocol was approved by human subjects committees at the National Institute of Public Health in Mexico and the National Institute of Environmental Health Sciences in the U.S. Results Among the 20 subjects in the standardization exercise, the between-subject coefficient of variation was greater for measures of AGD in females than for the other measures (Table 1 ). The reliability of the AGD measures were lower than for the traditional measures of anthropometry, with the female value being slightly lower than that for males. The variances estimated from the ANOVA were, for females: between-subjects, 7.9; between-observers, 0.6; and within-observer, 0.0. For males, the values were: between-subjects, 3.5; between-observers, 0.0; and within-observer, 0.1. The relative size of the variance components was unchanged when birth weight was included in the models. Thus, the between-subject variation in AGD was much greater than the variation due to measurement error. Table 1 Mean, coefficient of variation (CV), and reliability of anthropometric measurements in 20 newborns a Measurement Mean CV Reliability Weight (kg) 3.01 0.13 1.00 Length (cm) 48.9 0.03 0.97 Head Circumference (cm) 34.2 0.03 0.98 Anogenital distance 18 0.31 0.91 Female 11 0.27 0.50 Male 21 0.09 0.64 a 7 females and 13 males. Among the 87 subjects in the main study, the birth weight, length, and head circumference were as expected in a population from southern Mexico (Table 2 ) [ 27 ]. The AGD measure was about two-fold greater in males than in females, and there was little overlap in the distributions for males and females (Figure 2 ). The correlation of AGD with body weight was 0.64 in females and 0.48 in males. Table 2 Distribution of selected characteristics in 87 newborns, Mexico, 1999 a Variable Female n = 42 Male n = 45 Anogenital distance (mm) Mean 11 21 SD 2 3 Median 11 22 25 th percentile 10 20 75 th percentile 11 23 Weight (g) Mean 3070 3060 SD 408 440 Median 3060 3110 25 th percentile 2870 2800 75 th percentile 3310 3290 Length (cm) Mean 48.6 48.7 SD 1.4 2.2 Median 48.6 48.7 25 th percentile 47.5 48.0 75 th percentile 49.6 49.9 Head circumference (cm) Mean 337 341 SD 10.9 16.7 Median 337 341 25 th percentile 330 334 75 th percentile 345 350 a SD, standard deviation Figure 2 Distribution of Anogenital Distance (AGD), by Sex In the crude models of AGD in females, weight, length, and gestational age all appeared to be predictive (Table 3 ). The adjusted results, however, suggested that weight of the newborn was the most important correlate, based on the p value being lower than for length or gestational age. For males, weight and length were more important than gestational age as determinants, and this pattern was seen also in the adjusted results (Table 4 ). Length had a slightly larger R 2 and slightly lower p value, suggesting it may be a marginally better predictor than weight in males. In a model of data for males that included weight, length, and gestational age, the p values for both length and gestation were less than 0.05, although the coefficient for gestation was negative. In a model of AGD based on data for males and females combined (results not shown), after adjustment for weight, the term for sex was clearly important (β for males = 10.9 mm, standard error = 0.4, p < 0.0001; change in R 2 due to addition of sex to model = 0.86). Table 3 Regression coefficients for anogenital distance as a function of characteristics at birth, females a Variable Crude Adjusted b Coefficient 95% CI p value R 2 Coefficient 95% CI p value R 2 Birth weight 0.002 0.002 0.003 0.000 0.41 0.002 c 0.001 0.003 0.000 0.43 Birth length 0.319 -0.005 0.642 0.061 0.09 0.141 c -0.189 0.471 0.407 0.22 Gestational age 1.296 0.516 2.076 0.002 0.21 0.501 d -0.282 1.283 0.217 0.43 a Units for regression coefficients are mm of AGD per unit characteristic (g, cm, or weeks). Results based on 42 females. CI, confidence interval. b Multivariate adjusted regression coefficients (adjustment factors listed below). c Adjusted for gestational age. d Adjusted for weight of newborn infant. Table 4 Regression coefficients for anogenital distance as a function of characteristics at birth, males a Variable Crude Adjusted b Coefficient 95% CI p value R 2 Coefficient 95% CI p value R 2 Birth weight 0.003 0.001 0.005 0.001 0.23 0.004 c 0.002 0.006 0.001 0.27 Birth length 0.671 0.348 0.995 0.000 0.28 0.914 c 0.499 1.329 0.000 0.33 Gestational age 0.356 -0.258 0.971 0.262 0.03 -0.560 d -1.284 0.165 0.137 0.27 a Units for regression coefficients are mm of AGD per unit characteristic (g, cm, or weeks). Results based on 45 males. CI, confidence interval. b Multivariate adjusted regression coefficients (adjustment factors listed below). c Adjusted for gestational age. d Adjusted for weight of newborn infant. Discussion The AGD measures employed in the present study reflect the location of the caudal border of the genital swelling, an embryologic structure that differentiates into the labia majora in females and the scrotum in males. After the indifferent stage of the external genitalia, the critical events determining the sexual dimorphism of AGD in humans begin when, relative to the anus, the genital swelling, urethral folds, and possibly the genital tubercle, move ventrally under the influence of androgens [ 28 ]. Elongation of the genital tubercle, which becomes the phallus, also occurs at this time. The difference between males and females in our data demonstrates sexual dimorphism of this particular measure of AGD. The two-fold difference in the aspect of AGD that we measured is not reflected in the schematic diagrams of human sexual differentiation we have seen [ 29 , 30 ], which is likely due to the previous lack of formal measures. Direct comparison of our results with those in the two other studies with measures of anus-to-fourchette (AF) distance in female newborns [ 24 , 25 ] is hampered by different eligibility criteria, and possibly different ethnicities, in the three studies. For example, Callegari et al.'s subjects had a mean weight of 2,530 g; Phillips et al. did not present mean birth weight but subjects were required to have a birth weight above 2,750 g; and in our study the mean birth weight among females was 3,060 g. The mean AF distance in the Callegari et al. study was 10.9 mm; in the Phillips et al. study was 16.1 mm in Jews and 16.5 in Bedouins, and in the present study was 11 mm. Callegari reported no ethnic differences in their population (62.6% Hispanic, 28.7% black, and 8.7% white). Despite the ethnic-specific mean values noted above, Phillips et al. reported that Jewish females had a greater AF distance than did Bedouins. The similarity of the mean AF distance measures in the present study and the Callegari et al. study is surprising given the difference in mean birth weights, and suggests ethnic differences, or a systematic difference in how the measurements were done. Compared with established anthropometric measures on newborns, the reliability of the AGD measures were lower. The lower reliability of the AGD measures is likely due to several factors. The AGD measures depend on indistinct landmarks on soft tissues. Structures such as "the center of the anus" or the posterior fourchette are not clearly demarcated. Any slight traction or pressure applied to the perineum or surrounding structures could alter measures. Finally, compared with established anthropometric measures on newborns, the AGD dimensions are smaller, thus measures done with the naked eye on a subject unlikely to hold still are inherently at a disadvantage. Use of two observers, one to restrain the subject and one to do the measurements could result in improved reliability compared to our approach, which employed one observer. Compared with adult humans, the size of the genitals at birth is large relative to the body overall [ 28 ]. Yet the genital size is, of course, still determined in part by overall body dimensions and age. The need to adjust AGD for overall body dimension is well known in animal experiments [ 19 ]. In humans, the best approach to such adjustment remains unclear. Our data suggest that for the aspect of AGD we measured, adjustment for body weight is reasonable. A complete assessment of AGD in humans would include more measurements than were done in our study. In neonatal rodents, measurement of AGD is relatively straightforward and is the distance from the genital tubercle to the anus. In older animals or humans of any age, however, questions arise as to which measure is most informative. For example, in human males, rather than a genital tubercle, the presence of the phallus and testicles at birth means that a number of measurements are possible. The measurement in the present study, from the posterior scrotal-perineal junction, represents only one such measurement. Ideally we would have done genital tubercle measurements in males and females, but we did not. Whether sexual dimorphism exists in the distance from the anus to the genital tubule (penile base in males) would be useful to know. While one might expect that penile length may be a good measure of androgenization among males, difficulties obtaining a reliable measure mean that alternative measures, such as AGD, are worth investigating. Effects of endocrinopathies on AGD in humans have been described, but only to a limited degree. A rare form of congenital adrenal hyperplasia that causes incomplete masculine development has been reported to cause decreased AGD in boys [ 21 ]. Details on how the measurement was done (and the measured values), however, were not presented [ 22 , 23 ]. Callegari et al. [ 24 ] measured the distance from the anus to the fourchette (same as what we did) and in addition measured the distance from the anus to the clitoris; the ratio of these two measures in three newborn females with congenital adrenal hyperplasia was increased relative to normal newborn females. Earlier case reports on females with adrenogenital syndrome noted labiosacral fusion, but again, no formal measures were published [ 23 ]. The utility of AGD measures in humans is further supported by experimental data in primates showing that in utero exposure of females to androgenic agents increased AGD [ 1 ]. The purported mechanism by which androgens increase AGD in females is by inducing "labioscrotal fusion" (in normal males fusion begins caudally and proceeds ventrally, presumably androgens in females act the same way) [ 24 ]. This mechanism, however, does not account for why males who are not fully androgenized would have a decreased AGD, unless AGD in males is defined as being from tip of penis to the center of the anus. A set of formal AGD measures on subjects with selected congenital endocrinopathies or birth defects could be useful in evaluating whether this outcome is uniformly responsive to gross stimuli, and may help discern details of normal embryology and the consequences of disrupting it. Conclusions In summary, we have shown that an aspect of genital dimension that reflects migration of the genital swelling is sexually dimorphic in humans. Whether this particular measure, or other measures of AGD in humans, has any utility as markers of exposure in utero to hormonally active agents remains to be seen. Abbreviations AF: anus-fourchette AGD: anogential distance ANOVA: analysis of variance CI: confidence interval Competing interests None declared. Authors' contributions ES participated in the design of the study, carried out the measurements, and wrote the first draft of the manuscript. PR participated in the study coordination and data management. EY carried out and coordinated the measurements. ML originated the idea that AGD measurements in human males may be useful, revised the manuscript, and analyzed the data. MH conceived of the study and participated in its design and coordination. All authors read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC521084.xml
555739	Desmoplastic small round cell tumour: Cytological and immunocytochemical features	Background Desmoplastic small round cell tumor (DSRCT) is a rare and highly aggressive neoplasm. The cytological diagnosis of these tumors can be difficult because they show morphological features quite similar to other small round blue cells tumors. We described four cases of DSRCT with cytological sampling: one obtained by fine needle aspiration biopsy (FNAB) and three from serous effusions. The corresponding immunocytochemical panel was also reviewed. Methods Papanicolaou stained samples from FNAB and effusions were morphologically described. Immunoreaction with WT1 antibody was performed in all cytological samples. An immunohistochemical panel including the following antibodies was performed in the corresponding biopsies: 34BE12, AE1/AE3, Chromogranin A, CK20, CK7, CK8, Desmin, EMA, NSE, Vimentin and WT1. Results The smears showed high cellularity with minor size alteration. Nuclei were round to oval, some of them with inconspicuous nucleoli. Tumor cells are clustered, showing rosette-like feature. Tumor cells in effusions and FNA were positive to WT1 in 3 of 4 cytology specimens (2 out 3 effusions and one FNA). Immunohistochemical reactions for vimentin, NSE, AE1/AE3 and WT1 were positive in all cases in tissue sections. Conclusion The use of an adjunct immunocytochemical panel coupled with the cytomorphological characteristics allows the diagnosis of DSRCT in cytological specimens.	Introduction Desmoplastic small round cell tumor (DSRCT) is a rare and highly aggressive neoplasm described as a distinct clinicopathologic entity in 1989 by Gerald and Rosai [ 1 ]. Usually affecting young males and presenting as an abdominal mass, the tumor grows along serosal membranes with multiple nodules attached to the peritoneal surface [ 2 ]. Other primary sites have been reported as pleura [ 3 ], paratesticular region [ 4 ], bone and soft tissues [ 5 ] and ovary [ 6 , 7 ]. Histologically, a typical feature of DSRCT is the presence of clusters of tumor cells distributed within a cellular stroma. The shape of clusters varies from round to elongate. Tumor cells are small to medium-sized with round to oval hyperchromatic nuclei, with inconspicuous nucleoli. Necrotic cells and mitosis are common features. Cytoplasm is usually scanty, and cell borders are indistinct. Intracytoplasmic eosinophilic rhabdoid inclusions may be found in larger cells with nuclear pleomorphism [ 8 ]. The immunohistochemical profile shows divergent differentiation, a striking feature of this tumor. DSRCT may present a problem in the differential diagnosis with other round cell tumors. Tumor cells are immunoreactive for epithelial, neural and myogenic markers [ 2 ]. Cytogenetical studies have demonstrated a reciprocal chromosome translocation between the Ewing's sarcoma gene on chromosome 22 and the Wilms' tumour gene WT1 on chromosome 11, which is distinct from the translocation observed in Ewing sarcoma/peripheral neuroectodermal tumor (PNET) [ 9 ]. The cytological smears of DSRCT obtained by FNAB are moderately cellular. Tumor cells show round to oval nuclei with fine chromatin and inconspicuous nucleoli. Cytoplasm is scanty to moderate, with variable number of vacuoles. Tumor cells are arranged in loose clusters. Occasionally, spindle fibroblast-like cells are observed. Stromal fragments may be detected [ 10 ]. Effusion samples show cohesive cell clusters and similar cytological features. Mitoses or individual necrotic cells may be present, as nuclear molding [ 3 ]. In the current study, we describe the morphological and immunocytochemical features of four cytologic specimens, one of them obtained by FNAB and three from serous effusions (2 peritoneal fluid samples and one pleural effusion), from 3 patients with a diagnosis of DSRCT. Materials and methods We retrieved from the cytological files of Hospital do Cancer – A. C. Camargo four cytological specimens from 3 patients diagnosed with DSRCT, including one fine-needle aspiration sample and 3 fluid samples, during the last five years (2000–2004). FNA was performed on an inguinal mass of one patient. Alcohol-fixed smears were stained with Papanicolaou technique. Serous effusions were prepared with Cytospin (Shandon, Pittsburgh, Pennsylvania, USA). We evaluated two peritoneal fluid samples and one pleural fluid sample. One case (patient 1, peritoneal fluid) had a cellblock available. All cases were confirmed by histological analysis and immunohistochemical reactions. The histological sections were cut in sections of 4 μm and stained with H&E and immunohistochemistry. Immunocytochemical study was also performed on all cases. Immunohistochemical and immunocytochemical reactions were performed using streptavidin-biotin peroxidase technique with positive and negative controls. Diaminobenzidine was the chromogen. Table 1 shows the antibodies used and dilutions. All antibodies were from DAKO Corporation, Capinteria, CA, U.S.A. Table 1 Antibodies and dilutions used in this study Marker Antibody clone Dilution 34BE12 34BE12 1:100 AE1/AE3 AE1/AE3 1:500 Chromogranin A DAK-A3 1:100 CK20 KS20.8 1:50 CK7 OV-TL 12/30 1:100 CK8 35BH11 1:100 Desmin D33 1:100 EMA E29 1:2000 NSE BBS/NC/V1-H14 1:1500 Vimentin Vim 3B4 1:200 WT1 6F-H2 1:400 Cases Patient 1 22-year-old white female, with abdominal pain. Video-laparoscopy showed a liver mass and multiple peritoneal implants diagnosed as DSRCT. Six months after the diagnosis, she started chemotherapy for four months, and reduction of tumor mass was observed. One month after the end of chemotherapy, the tumor was removed. Macroscopically, tumor mass measured 5.0 × 4.0 × 3.8 cm and was involving uterus, pericolic tissue, and vagina. Histological analysis shows also involvement of both ovaries and large bowel wall. Ten out of 13 lymph nodes showed metastasis of DSRCT. The peritoneal fluid colleted during surgery was negative for neoplastic cells. Eight months after the first surgery, she presented with a recurrence in the abdominal cavity and a new resection of the tumor mass showed involvement of cecal appendix. Peritoneal fluid sample collected at that time was positive for malignant cells. In the follow up examination, seven months after the second surgery, it was found an inguinal tumor mass of 15 mm. FNA was performed and showed DSRCT metastasis. After the diagnosis, this patient was transferred to another institution. Patient 2 Seven year-old male with back pain and fever. CT scan showed pleural effusion and a mediastinal mass measuring 16.0 × 9.0 cm. Tumour mass showed involvement of soft tissues. Surgical biopsy and pleural drainage were performed. The patient was treated with radiotherapy and chemotherapy, but died 8 months after the diagnosis. Patient 3 Male, 15-year-old had acute abdominal pain and was submitted to an exploratory laparotomy that disclosed a large pelvic mass, involving epiplon and sigmoid, cecum, liver and peri-aortic lymph nodes. This patient had multiple nodules on peritoneal surface. The biopsy of tumor was performed. One month after the diagnosis, chemotherapy was initiated. The patient was submitted to chemotherapy during 8 months, with reduction of more than 50% of tumor mass. A second laparotomy was done to excise retroperitoneal and retrovesical mass. At this time peritoneal fluid sample was collected. After surgery, chemotherapy was continued. The patient is alive, with residual disease. Results Cytological findings Case 1 (Fine needle aspiration) The smears showed high cellularity. The tumor cells exhibited a slight variation in size. Nuclei were round to oval, some of them with small nucleoli. The cytoplasm was scanty. Tumor cells are clustered, with rare clusters showing rosette-like features. The background of the smears showed lymphocytes. (Figure 1 ). Figure 1 Clusters of small round tumor cells showing rosette-like features in smear of fine needle aspiration specimen of DSRCT. Cases 1, 2 and 3 All fluid samples showed similar features. The samples showed high cellularity. Tumor cells were more frequently arranged in tridimentional clusters, but occasionally, isolated cells are also seen. Additionally, clusters showing rosette-like features are rarely observed. Nuclei were round to oval, some of them with small nucleoli. The cytoplasm was scanty (Figure 2 ). Figure 2 Effusion from patient with DSRCT exhibiting high cellularity. Observe tridimentional clusters of neoplastic cells. Nuclei were round to oval, some of them with small nucleoli and the cytoplasm is scanty. Immunohistochemical and Immunocytochemical findings The distribution of immunoreactivity in histological and cytological samples from the patients are summarized in Table 2 . Tumor cells in effusions from patients 1 and 2 and, the smear obtained by FNA (Patient 1) were positive to WT1 (Figure 3 ). Table 2 Distribution of immunoreactions in patients 1, 2 and 3 histological and cytological samples. Marker Patient 1 Biopsy Patient 2 Biopsy Patient 3 Biopsy Patient 1 Cytology Patient 2 Cytology Patient 3 Cytology 34BE12 - - ND ND ND ND AE1/AE3 + + + ND ND ND Chromogranin A + + ND ND ND ND CK20 - - ND ND ND ND CK7 - - ND ND ND ND CK8 + - ND ND ND ND Desmin + - ND ND ND ND EMA + + ND ND ND ND NSE + + + ND ND ND Vimentin + + + ND ND ND WT1 + + + + + - (Effusion & FNA) N.D. = not done. Figure 3 DSRCT tumor cells in effusion showing nuclear positive reaction to WT1. Patient 1 the histological sample collected before chemotherapy, exhibited immunohistochemical positivity for vimentin, epithelial membrane antigen (EMA), neuron specific enolase (NSE), chomogranin A, and desmin in dot-like perinuclear pattern. Cytokeratin expression was observed with anti-cytokeratin cocktail (AE1/AE3) and Cytokeratin 8. Tumor cells also expressed WT1 protein. Patient 2 tumor cells exhibited positivity for vimentin, EMA, NSE, chomogranin A, AE1/AE3 and WT1. Desmin and cytokeratins 7, 20 and 34BE12 were negative. Patient 3 tumor cells exhibited positivity for vimentin, NSE, AE1/AE3 and WT1. The immunohistochemical study was performed before chemotherapy. Discussion DSRCT is a rare neoplasm that affects young patients. It may present a problem in the differential diagnosis with other small round cell tumors. The diagnosis of DSRCT however can be established with correlation of clinical, cytological and immunocytochemical features. The cytological features that we found in the smears obtained by FNA are similar to other descriptions in the literature. Similar to reports of Zeppa et al [ 11 ], we did not detect in our smears fragments of fibrosis or cytoplasmic granules or vacuoles. The finding of stromal fragments, frequently seen in FNA is not a common finding in liquid based preparations [ 12 ]. One of the characteristics of DSRCTs is its dissemination along serous surfaces. Due to this fact, development of serous effusions is a common clinical finding in DSRCTs patients, with detection of tumor cells in the fluid. In effusions, tumor cells may be present in aggregates but no obviously architectural arrangement is seem. Demonstration of a divergent phenotype and the reciprocal translocation characteristic of DSRCT are critical to the diagnosis. In a reported series of 32 cases of DSRCTs [ 13 ], 88% of cases were immunoreactive for AE1/AE3, 84% for NSE, 81% for desmin. These results were similar to other previous studies [ 2 ]. Lae et al [ 13 ] detected positivity to WT1 antibody in 29 out of 32 cases. Our immunohistochemical results are in agreement with other previous studies. Strong membrane expression of HER2/neu and immunoreactivity to c-kit protein are not common findings [ 14 ]. The establishment of a specific reciprocal translocation t (11; 22)(p13;12) as diagnostic in DSRCT was based on the results of Sawyer et al [ 9 ]. Shen et al [ 15 ] and Roberts et al [ 16 ] described variants of with other chromosome involved in addition to chromosome 11 and 22. The translocation t (11; 22)(p13;12) involve the EWS gene in 22q24 and WT1 gene in 11p13. This translocation produces the chimeric transcript EWS/WT1 and the related WT1 protein, which can be detected by immunohistochemical method. EWS gene encodes a protein which the precise function and normal role has not yet been elucidated. Recently, Thomas et al [ 17 ] proposed that the protein product of the EWS gene interacts with Brn-3a cellular transcription factor via a direct protein-protein interaction. Native WT1 protein function has not completely known, but it represses transcription in vitro of many genes. WT1 is a tumor-suppressor gene that encodes a protein, which mediates transcriptional repression and interacts with p53 protein [ 18 ], product of another tumor suppressor gene, TP53, frequently deleted or mutated in many human tumors. In absence of intact p53 protein, WT1 acts as a transcriptional activator [ 19 ]. Normal WT1 protein is expressed in tissues, which undergo mesenchymal-epithelial conversion derived from mesoderm, in a specific period of development [ 20 ] and it plays a role in mesothelial formation in embryonic development [ 21 ]. Immunohistochemical detection of WT1 in DSRCTs is predictive of the translocation and it also demonstrates that the chimeric protein is expressed in significant amount in tumour cells 22, 23 . In addiction to consistent WT1 expression, the typical serosal involvement in DSRCT has raised the possibility that this tumor might be a blastematous tumour derived of primitive mesothelium [ 24 ]. Mesothelin is a glycoprotein of unknown function strongly expressed in mesothelial cells. Although lack of specificity of expression of mesothelin for mesothelial origin, the expression of this protein in DSRCT may have some significance on histogenisis of this tumor [ 25 ]. We detected WT1 immunoreactivity in all tumors tissues and in 2 out of 3 serous effusions with malignant cells, as well as on FNAB smears. The high frequency of DSRCTs with WT1 protein expression suggests that in consensus with clinical tomographic and cytological findings, this antibody may be used to confirm the diagnosis of DSRCT in cytological samples. We observed a negative WT1 reaction in the cytological sample of patient 3. This sample was collected 10 months after the end of chemotherapy protocol. We can hypothesize if chemotherapy hampered a different antigenic pattern in malignant cells, and influenced this result. Among other small round cell tumors, most of cases of rhabdomyosarcomas and neuroblastomas do not disclose nuclear WT1 staining [ 26 , 27 ]. Comparing DSRCT and Ewing Sarcoma/PNET, Hill et al. [ 28 ] detected WT1 nuclear immunoreactivity in all 13 DSRCT cases studied; conversely, all 11 cases of Ewing Sarcoma/PNET were negative. Additionally, Wilm's tumor was demonstrated to present a high percentage of cases with nuclear WT1 staining; for this reason, correlation with clinical findings is necessary to do a differential diagnosis between Wilm's tumour and DSCRT in effusions [ 26 ]. On the other hand, it is important to emphasize that malignant mesothelioma should also be considered in the differential diagnosis, since it can show varied histological appearances including sarcomatoid differentiation with desmoplastic areas, or even resembling undifferentiated sarcomas [ 29 ]. WT1 might also decorate nuclei of both epithelioid or biphasic mesothelioma but in general, WT1 stain most frequently epithelioid mesotheliomas [ 30 ]. The use of a panel of markers can also help in the differential diagnosis. In conclusion, cytological and immunophenotypical findings in an appropriate clinical context is sufficient to suggest DRSTC, what sounds highly contributive for us, considering the high aggressiveness of this tumor.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC555739.xml
406407	Open Access and Scientific Societies	Societies are encouraged to consider their own open-access experiments within the context of the communities they serve	This is the second in a series of three editorials that aim to address recurring concerns about the benefits and risks associated with open-access publishing in medicine and the biological sciences. Scientific societies serve their members, their broader scholarly communities, and the different components of their missions in many important ways. Making peer-reviewed literature immediately accessible, searchable, and reusable to anyone in the world with an Internet connection is a uniquely direct means of achieving a number of goals that are common to most scholarly associations and of advancing the diverse interests of their constituencies. Setting aside for the moment the question of how feasible it is for societies to alter their journals' access policies, there is by now a broad consensus that widespread open access to scientific publications is good for scientists and good for science. Society members want to maximize the impact of their work—and articles that are freely available online are cited more frequently than those that are not ( Lawrence 2001 ). Most societies are committed to catalyzing innovations within and across scientific disciplines—and open-access archives of full-text literature provide a valuable tool for sharing information globally in order to accelerate the rate of scientific progress. Many societies articulate in their mission statements the goal of communicating the benefits of their members' discoveries with the public—and open-access publishing is a direct means to accomplish this goal. In addition to an interest in exploring new ways to serve their members and their missions, societies have another compelling reason to investigate open access for their journals: the rapidly changing landscape of scholarly publishing. From 1990 to 2000, the average price of an academic journal subscription increased 10% per year ( Create Change 2000 ). While society-run and nonprofit journals may not be the major contributors to those spiraling costs, societies that rely on revenues from subscriptions and site licenses may bear a disproportionate share of the negative consequences of skyrocketing serials prices. As libraries are forced for a variety of reasons (including decreased budgets and the increasing prevalence of “big deals” and journal bundling) to eliminate subscriptions, society journals may be among the hardest hit. Journals that appeal to a relatively specialized readership and those that are not part of larger publishing groups are particularly vulnerable to the contraction of serials collections that has already begun and will likely accelerate ( Create Change 2000 ). A Society Is More Than a Journal The confluence of forces in favor of open access says nothing about its fiscal implications for scientific societies. As any systemic change in research or publishing would, the movement toward open access has generated concern about its ramifications for the scholarly associations that often serve as the backbones of scientific communities. However, the strength of those societies and their essential role in the communities they serve are precisely what should allay fears about the revenue-eroding effect that some argue would plague societies if they converted their traditional subscription-based journals to open access. Scientific societies perform an array of tremendously valuable functions for their constituents and disciplines. Researchers, educators, and others join societies for the many benefits of membership beyond simply discounted or “free” subscriptions to journals, so the concern that open-access publications would be the death knell of voluntary academic associations is misguided. As Elizabeth Marincola, executive director of the American Society for Cell Biology, recently noted, her society “offers a diverse range of products so that if publications were at risk financially, we wouldn't lose our membership base because there are lots of other reasons why people are members” ( Anonymous 2003 ). While open-access publication can, in fact, be paid for in a number of different ways, there is no question that a transition toward the elimination of online access barriers requires most societies to restructure the business models for their journals. If journal subscriptions generate surplus revenue that supports other society activities, then the business model of the society as a whole may need to be examined. This is not to say that open-access journals cannot generate a surplus or profit—simply that they do not do so by restricting access to their primary research content. Testing the Open-Access Waters There are a number of societies that have already begun to take transitional steps to wean themselves from subscription revenues. One of the earliest societies to commit to open-access publication, the American Society for Clinical Investigation (ASCI) has since 1996 provided the Journal of Clinical Investigation (JCI) freely online and recently reaffirmed its commitment to open access: “The financing having been resolved, through author charges and other means,” John Hawley, the executive director of the ASCI writes, “the JCI hopefully can bring the greatest benefit to its authors and readers, regardless of who they might be. It is in this spirit that the JCI has always been free online, and will remain so” ( Hawley 2003 ). In order to experiment cautiously with new access policies, several societies have implemented hybrid models of access-restriction for their publications. The American Physiological Society, for example, offers authors in Physiological Genomics the option to pay a surcharge for their articles to be made freely available online immediately upon publication. A recent survey by the Joint Information Systems Committee (JISC) in the United Kingdom suggests that many authors would use such an option if it were more widely available: 48% of authors who had never published in an open-access journal and 60% of authors who had done so indicated that they would be willing to “pay a publisher of a journal sold according to the traditional subscription model an additional fee for them to make [the author's] particular paper ‘open access’” ( JISC 2004 ). JISC is also directly encouraging society and nonprofit publishers to implement hybrid models and other open-access experiments and to launch new open-access journals by providing grants to offset the publication charges for authors during this transitional phase. In the long run, of course, open access will prove sustainable when more funders of research, in addition to interested third parties, designate funds specifically for the costs of publishing articles to be made freely available, searchable, and reusable online. Starting the Dialogue Reaching a “steady-state” system of open-access publishing by scientific societies will require three critical components: recognition that open access serves societies' members and missions; diversified revenue streams not solely dependent on subscription or site-license fees; and society publishers' making use of recent innovations in journal production and dissemination, which can dramatically reduce the costs of publishing. It is, after all, the increased efficiencies born of new technologies—from the Internet itself to electronic journal management systems—that have made the idea of open access possible. And while proponents of open access are confident that publication charges of around $1,500 per article will be sufficient to cover the costs of publishing an efficiently operated society journal, there is no question that many existing journals may need to update their infrastructure in order to make open access financially viable ( PLoS 2004 ). There is also no question that many societies do not, at present, have a wealth of revenue streams beyond the proceeds from their journals, which they often use to fund valuable activities from education initiatives to annual meetings. As open-access journals become more established, however, and as the benefits of open access to scientific and medical literature become more apparent to society members, the demand for the broadest possible dissemination of research is only likely to grow. Those societies that embrace the developments taking place in scholarly publishing may well see their membership and publications thrive more than societies that cling to the potentially unstable status quo. In any case, a constructive discussion about the pitfalls to be avoided and the benefits to be gained through a transition to open-access publishing would be a worthy first step for any scientific society to take—and PLoS welcomes the questions, comments, and feedback of those who are intrigued by the potential that open access affords and want to learn more.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC406407.xml
550669	An AFLP-based genetic linkage map of Plasmodium chabaudi chabaudi	Background Plasmodium chabaudi chabaudi can be considered as a rodent model of human malaria parasites in the genetic analysis of important characters such as drug resistance and immunity. Despite the availability of some genome sequence data, an extensive genetic linkage map is needed for mapping the genes involved in certain traits. Methods The inheritance of 672 Amplified Fragment Length Polymorphism (AFLP) markers from two parental clones (AS and AJ) of P. c. chabaudi was determined in 28 independent recombinant progeny clones. These, AFLP markers and 42 previously mapped Restriction Fragment Length Polymorphism (RFLP) markers (used as chromosomal anchors) were organized into linkage groups using Map Manager software. Results 614 AFLP markers formed linkage groups assigned to 10 of 14 chromosomes, and 12 other linkage groups not assigned to known chromosomes. The genetic length of the genome was estimated to be about 1676 centiMorgans (cM). The mean map unit size was estimated to be 13.7 kb/cM. This was slightly less then previous estimates for the human malaria parasite, Plasmodium falciparum Conclusion The P. c. chabaudi genetic linkage map presented here is the most extensive and highly resolved so far available for this species. It can be used in conjunction with the genome databases of P. c chabaudi , P. falciparum and Plasmodium yoelii to identify genes underlying important phenotypes such as drug resistance and strain-specific immunity.	Background Plasmodium chabaudi chabaudi is a malaria parasite of murine rodents. It has been widely used as a model to study various aspects of parasite biology and disease which are difficult to investigate using human malaria parasites. For instance, P. c. chabaudi is being used to study the genetic basis of drug resistance [ 1 - 4 ] and strain-specific immunity [ 5 ], because the execution and analysis of genetic crosses is relatively straightforward in this species [ 6 ]. The analysis of the genetic basis of aspects of malaria biology has been facilitated by recent developments in malaria genomics. Firstly, the Plasmodium falciparum genome has been fully sequenced and mapped [ 7 ] and there is also extensive sequence data now available for three of the four main malaria parasites of murine rodents [ 8 ]. Secondly, the degree of homology and conservation of gene synteny between the various species of malaria [ 4 , 9 , 10 ] allows the undertaking of comparative genomics and facilitates the elaboration of accurate genomic maps in these species. However, a genetic linkage map of the 14 chromosomes of P. c. chabaudi is still important for the identification of loci which influence phenotypes such as drug resistance. A previous genetic linkage map of P. c. chabaudi was generated using over 40 RFLP markers [ 11 ]. However, due to the small number of markers available, this linkage map had limited usefulness. The authors have recently developed a large number of genome-wide polymorphic AFLP markers for P. c. chabaudi [ 11 ]. AFLP markers have previously been used to generate genetic linkage maps in another apicomplexan parasite, Eimeria tenella [ 12 ], as well as in Trypanosoma brucei [ 13 ]. This article presents a high-resolution genetic linkage map of P. c. chabaudi and an estimate of map unit size. The value of the genetic linkage map in the identification of genes determining selectable phenotypes is also described. Methods Mouse strains used in experiments Inbred female CBA mice, obtained from the University of Edinburgh, were used for the growth of P. chabaudi parasites. Mice were housed in propylene cages with sawdust bedding and were fed on Harlan, SDS formula number I (Special Diet Services Ltd.) and drinking water supplemented with 0.05% paraminobenzoic acid (PABA) to aid parasite growth [ 14 ]. Temperature was maintained between 22 and 25°C with a 12 hour light/12 hour dark cycle. Parasite lines Clones of the genetically distinct isolates AS and AJ, originally isolated from wild thicket rats, Thamnomy rutilans [ 15 ] were used as parents in genetic crosses. 28 recombinant clones were analysed here. 20 clones originated from a cross between AJ and AS (3CQ) (a chloroquine-resistant clone derived from AS) [ 1 ], while 8 clones originated from a cross between AJ and AS (30CQ) (a clone with higher resistance to chloroquine, derived from AS (3CQ) [ 16 ]. Maintenance of parasites For routine maintenance of parasites, parasitized red blood cells collected from the tail veins of infected mice were passaged in citrate saline into uninfected mice. Cryopreservation of infected blood was performed by exsanguination of mice anaesthetised with halothane. Blood was collected in a tube containing 2–3 volumes of citrate saline (0.9% NaCl, 1.5% tri sodium citrate dihydrate, adjusted to pH 7.2). The mixture was spun at 2000 rpm for five minutes, the supernatant discarded and the red cell pellet mixed with two volumes of a solution containing 28% (v/v) glycerol, 3% sorbitol and 0.65% NaCl. The mixture was then aliquoted into several glass capillaries, which were sealed by flame and deep-frozen in liquid nitrogen. 50 μl of cryopreserved blood was recovered by thawing capillaries into 10 μl of 12% NaCl and mixing for 3–5 mins. Nine volumes of 1.6% NaCl were then added dropwise and samples centrifuged at 2000 rpm for 3–5 mins. The supernatant was removed and nine volumes of 0.9% NaCl/ 0.2% dextrose solution were added dropwise. After mixing, the mixture was centrifuged again, supernatant removed and red blood cells resuspended in a 0.9% NaCl/ 0.2% dextrose solution for injection. Estimation of parasitaemia Parasitaemia was estimated by microscopic observation of thin blood smears taken 4–6 days after parasite injection and stained with 20% Giemsa staining solution (BDH) for 15 minutes. Parasitaemia was estimated by calculating the percentage of red blood cells infected in at least five microscopic fields. Preparation of parasite DNA Each parasite DNA preparation was obtained from five infected CBA female mice. Blood samples were taken from mice infected with AS, AJ and recombinant clones having high parasitaemias in the mid-afternoon, when parasites were trophozoites. Host lymphocytes or nucleated cells present in the blood were removed as described previously [ 11 ]. Parasites were pelleted and stored at -70°C. DNA was extracted and purified as previously described [ 11 ] and stored at -20° for future use. Amplified Fragment Length Polymorphism (AFLP) technique The AFLP method was carried out according to the original protocol [ 17 ] with slight modifications, as described by Grech et al [ 11 ]. Briefly, parasite genomic DNA was digested with two enzymes, Eco RI and Mse I, and ligated with adapters, to provide the complementary sequences for AFLP primers. The first round of amplification used primers containing the Eco RI or Mse I recognition sequences at their 3' end. The second round of (selective) amplification use two additional (selective) bases (3' terminus) in both primers, one of which (the Eco RI primer) was radiolabeled with γ-[ 33 P] ATP. PCR products were run on acrylamide gels and AFLP bands visualised on autoradiography films. Polymorphic bands between the two parental strains were used as markers for the genetic linkage map. Organization of AFLP markers in a genetic linkage map For every marker, the parental alleles identified in each of the progeny clones were entered in an Excel spreadsheet. The absence of a band in one parent was treated as the presence of the other parental allele at that locus. Data were then prepared for analysis with the Map Manager QTX software [ 18 ] according to the instruction manual. The dataset was designated as "Backcross" for the purpose of computer analysis. Prior to linkage analysis, markers were tested for random assortment using a chi-square test, to exclude markers segregating in a non-random fashion from the initial analysis with Map Manager, and thus to avoid spurious linkage inferences between such markers. Because of the large number of tests for non -random assortment performed (n = 672), some markers showing apparent non-random assortment may have been falsely excluded from our analysis; i.e. some valid markers may indeed show non-random assortment and should be included. A Bonferroni correction was therefore applied to the chi-square test to decrease the stringency of the statistical test. The value of the Bonferroni correction represents the number of 'independent' comparisons and here was arbitrarily set at 24. This value was chosen as representing the likely number of chromosomal fragments at meiosis, and is supported by data presented in this paper. Markers within these fragments are not independently inherited. The chosen value (24) is a compromise between 672 (which assumes that all 672 markers are independently inherited) and 14 (the number of chromosomes, and which assumes that no pair of markers on one chromosome are inherited independently). Markers not following random assortment in the initial test were thus divided into two groups, i.e. those segregating in a non-random fashion before and after Bonferroni correction, and those segregating in a non-random fashion before but not after the Bonferroni correction. The markers in the latter group were added separately after linkage groups had been determined (see below). Linkage groups using AFLP and RFLP markers were formed with an initial p-value of 0.0001 using the "Make Linkage Groups" command in Map Manager. p-values in Map Manager indicate the probability of a Type 1 error; that is, the probability of a false positive linkage. Following formation of linkage groups, the p-value was raised to 0.001. Linkage at p < 0.001 is considered significant. Using the command "Distribute", linkage groups were brought together. Then, other previously unlinked markers were allocated to these new linkage groups, again using the "Distribute" command. Markers with non-random assortment after statistical analysis without Bonferroni correction were added next, and those still segregating in a non-random fashion after Bonferroni correction were added last. The "Ripple" function was then used to position markers in an order which maximizes the total LOD (logarithmic odds) score for linkage. The software also estimated the optimum order and genetic distance between markers in centiMorgans (cM) by using the "Kosambi" function in the software. It was then possible to calculate a map unit size (i.e the physical distance corresponding to 1 cM). The presence of 42 previously characterised RFLP markers which had been physically mapped onto P. c. chabaudi chromosomes [ 1 ] served as anchors for the placement of AFLP linkage groups onto specific chromosomes. Results The inheritance of 672 AFLP markers was determined in 28 progeny clones derived from two crosses between P. c. chabaudi AJ and either clone AS (3CQ) or AS (30CQ). The majority of the AFLP markers showed independent assortment in the 28 progeny clones, as illustrated previously [ 11 ]. However, 66 markers failed the chi-square test at 5%, 15 of which failed it after the Bonferroni correction. Markers were allocated to linkage groups using the Map Manager program and groups assigned to chromosomes using 42 previously mapped RFLP markers as anchors [ 1 ]. Estimated numbers of recombination events, genetic lengths of chromosomes and recombination frequencies were also determined for the identified chromosomes using Map Manager. Allocation of markers to linkage groups The 672 AFLP markers formed a total of 22 linkage groups with a final p-value of 0.001. Additional file 1 summarises the numbers of AFLP and RFLP markers assigned to each chromosome or to unassigned linkage groups, the estimated physical size of each chromosome [ 19 ] and the number of AFLP markers per Mb. 400 AFLP markers in 10 linkage groups could be assigned to P. c. chabaudi chromosomes 1 and 5–13, by virtue of their linkage to RFLP markers previously assigned to specific chromosomes by physical mapping [ 1 ]. 272 AFLP markers could not be assigned to a specific chromosome. 214 were placed in 12 unassigned linkage groups, each with between 2 and 51 AFLP markers. At least four of these linkage groups are likely to map to chromosomes 2, 3, 4 or 14. The failure to assign these linkage groups occurred because RFLP markers previously mapping to chromosomes 2, 3, 4 and 14 were not allocated to linkage groups. This was probably due to insufficient characterization of the inheritance patterns of these RFLP anchors which were determined in a small number of recombinant clones [ 1 ]. For instance, the inheritance of a RFLP marker assigned to chromosome 2, Ca-ATPase, was only determined for 7 out of the 28 recombinant clones. No independent physical mapping of unassigned linkage groups was attempted here. 58 AFLP markers, 21 of which segregated in a non-random fashion, could not be allocated to any linkage groups. Physical mapping of these markers would be required to assign them to specific chromosomes. Alternatively, unassigned linkage groups or unallocated markers might map to the small mitochondrial or apicoplast genomes, although these combined represent only 0.2% of the genome. Several RFLP markers could not be allocated to linkage groups by the Map Manager software, probably due to the small numbers of clones analysed for these markers. These markers were added to assigned linkage groups according to their chromosomal assignment, as previously determined by physical mapping [ 1 ]. With the exception of chromosomes 2, 3, 4 and 14 (discussed above), chromosomes 9 and 10 showed the lowest density per Mb of AFLP markers. Chromosome 7 showed the highest density. This may simply reflect natural variation in the frequency of AFLP polymorphisms on particular chromosomes. However, for chromosomes with a low apparent density of AFLP markers such as chromosomes 9 and 10, it is likely that some of the markers in unassigned linkage groups would physically map to these chromosomes. These unassigned groups may not show genetic linkage with (groups of) assigned markers because of factors such as a high rate of recombination between two linkage groups (one linked to the RFLP anchor) or because of an intervening region with a low density of AFLP markers. Both factors, or a combination of the two, may prevent two physically linked groups from being identified as genetically linked. Conversely an apparent unusually high frequency of AFLP markers (as in chromosome 7) may arise from strong linkage disequilibrium between loci on two different chromosomes. Some markers located on one chromosome may thus appear to be genetically linked to markers on another. This could arise where one locus exerts a strong constraint on another unlinked locus. For instance, the AJ allele of an enzyme in a metabolic pathway may only function with the presence of the product of the AJ allele encoding another enzyme in the same pathway. This constraint might be structural or functional. The same may be true of AS alleles of the same enzymes. In this case, the genes encoding these enzymes, and markers strongly linked to them, may appear in the same genetic linkage group. Order of markers in the linkage groups AFLP markers were initially ordered on the linkage groups as described in Materials and Methods. After inspection of the predicted marker order, occasional manual adjustments were made to correct markers which appeared to be inappropriately positioned. Because Map Manager failed to allocate some RFLP markers to an assigned linkage group, these were positioned manually. The final distribution of the markers on the 10 linkage groups assigned to chromosomes is shown in Fig. 1 , 2 , 3 (See also Additional file 2 , Additional file 3 . and Additional file 4 ). 7 unassigned linkage groups containing 9 or more markers each are shown in Fig. 4 (see also Additional file 5 ). Figure 1 Linkage map for chromosomes 1, 5, 6 and 7 of the P. c. chabaudi genome. The AFLP and RFLP markers assigned to chromosomes are displayed with genetic distances (in cM). AFLP markers were named as follows: the first two letters identify the clone to which a marker is specific, the next two letters indicate the Eco RI primer selective bases, the numbers identify the marker for that clone and primer combination in order of its molecular size, and the last two letters identify the Mse I selective bases. RFLP markers were based on genes previously identified [1]. Figure 2 Linkage map for chromosomes 8–11 of the P. c. chabaudi genome. The AFLP and RFLP markers assigned to chromosomes are displayed with genetic distances (in cM). Figure 3 Linkage map for chromosomes 12 and 13 of the P. c. chabaudi genome. The AFLP and RFLP markers assigned to chromosomes are displayed with genetic distances (in cM). Figure 4 Unassigned linkage groups containing more than 8 markers. The AFLP markers assigned to unassigned linkage groups are displayed with genetic distances (in cM) Number of recombination events per chromosome If markers and recombination events were both uniformly distributed across the genome, then we would expect the number of predicted recombination events (totalled from 28 clones characterised here) in each chromosome to correlate with its physical size. The predicted total number of recombination events occuring in each linkage group is shown in the Additional file 1 . Uniformity was evaluated by comparison of the frequency of recombination events (from the 28 clones) in each chromosome. This varies between 6.7/Mb (chromosome 10)) and 31.1/Mb (chromosome 7), with an overall value of 13.3/Mb. Chromosomes 1 and 11 also showed low frequencies. These differences may reflect natural variation in recombination rates across the genome. However, other factors may also contribute. For instance, there is likely to be a systematic underestimation of recombination frequency because the physical extent of the linkage groups assigned to particular chromosomes will be less than their actual size. For instance, if the chromosome 10 linkage group extends across only half of the chromosome, then the real density of markers and recombination events (per Mb) is likely to be about twice the apparent value given. Indeed, Additional file 1 shows that when data from the unassigned linkage groups are included, the recombination frequency (across the whole genome) increases to 15.9 events per Mb. Regardless of the number of markers assigned to each chromosome, we would expect the number of recombination events per AFLP marker to remain relatively constant, if both the frequency of polymorphism and the rate of recombination vary little between chromosomes. This is indeed the case. For the different chromosomes, the measure varies only between 0.4 and 0.9 recombination events per AFLP marker (see Additional file 1 ). Assuming that the P. c. chabaudi genome consists of about 20 Mb [ 19 ] the value of 13.3 recombination events in 28 recombinant clones/Mb converts to about 9.5 recombination events/clone/genome which is very close to the value of 10 estimated for Plasmodium falciparum [ 21 ]. A significant number of double crossover events around a single AFLP marker were observed in many linkage groups. Any such occurrences were re-evaluated on the original X-ray film to detect any possible errors or ambiguous bands. Many distinct double crossover events were observed. The same phenomenon was also commonly observed in P. falciparum , and were interpreted as being due to non-reciprocal conversion events [ 20 ]. Genetic length of linkage groups The apparent genetic length of each linkage group in cM was calculated by Map Manager based on the number of recombination events ( Additional file 1 ). When added together, the linkage groups assigned to chromosomes combined to give a total genetic length of 1180 cM and the unassigned linkage groups a further 497 cM, giving a total for the genome of 1676 cM. Due to the limited number of clones available (28), the smallest genetic distance that could be measured between two markers was approximately 3.6 cM, corresponding to the presence of a single recombination event between two markers in 28 clones. In general, the estimated genetic lengths of the linkage groups assigned to chromosomes 1, 5–13 increased with the estimated physical sizes of the chromosomes (Figure 5 ). The Pearson correlation coefficient was 0.794 (p < 0.005). The estimated sizes of map unit for chromosomes 1 and 5–13 are shown in the Additional file 1 . These values varied from 8.9 kb/cM (chromosome 5) to 24.1 kb/cM (chromosome 11) with an overall estimated mean of 15.1 kb/cM. It is likely that there is some overestimation of physical size of a map unit because individual identified linkage groups are unlikely to cover the full extent of any chromosome. Indeed, when the genetic lengths of the unassigned linkage groups are included in the analysis, the overall map unit size is reduced to 13.7 kb/cM. The inclusion of additional unallocated markers may reduce this value even further. Figure 5 Relationship between genetic size and physical size of each chromosome. The relationship between the estimated genetic sizes and physical sizes of chromosomes 1, 5–13 (Table 1) is shown. The correlation between these two variables is 0.794. However, some overestimation of genetic length in linkage groups is also possible, which leads to an underestimation of map unit size. For instance, Figure 2 shows that chromosomes 9 and 10 both show two abnormally large sections bounded by RFLP anchors without intervening AFLP markers. Specifically, chromosome 9 shows, at one end, an RFLP marker, ran , 40 cM from its nearest AFLP marker. At its other end, an RFLP marker, Ag 3027, lies about 35 cM from its nearest AFLP marker. Chromosome 10 has RFLP marker cDNA121 about 70 cM from its nearest AFLP marker and RFLP marker, 5S rRNA, a further 70 cM distant. These large gaps may be artefacts which arise for two reasons. Firstly, some unreliability in the typing of clones using RFLPs was previously noticed [ 4 ]. Secondly, the inheritance patterns of these markers were not determined in all 28 recombinant clones. Markers ran , Ag3027, cDNA121 and 5S rRNA were typed for only 9, 17, 9 and 16 clones, respectively. The characterisation of inheritance of RFLP markers in all of the 28 progeny clones, and the correction of possible mistakes may reduce the estimated genetic length. This would lead to an increase in map unit size. Nevertheless, it is notable that the value reported above (13.7 kb/cM) is close to estimates made for P. falciparum (15–30 kb/cM [ 21 ] or 17 kb/cM [ 20 ]), although slightly smaller. It is likely that the recombination rate may vary within as well as between chromosomes or genomic loci [ 22 ]. Estimate of potential alleles due to indel mutations Of the 400 AFLP markers placed on chromosomes, 37 AS-AJ pairs (74 markers i.e. 18% of the total) shared the same selective bases at both primer ends and showed complementary segregation in the cross-progeny clones. Most of these markers also differed in size by only a few base pairs. They are likely to be alleles at the same loci. This was confirmed by sequencing two such pairs, namely ASTA01AC and AJTA01AC (chromosome 5), and ASTT02CA and AJTT02CA (chromosome 13) (sequence data not shown). This suggests that a significant proportion of the polymorphisms observed between AS and AJ may be due to small insertions or deletions. In fact, it was observed that small indels tend to occur in introns or intergenic regions (data not shown). Reliability of the AFLP markers in the progeny clones A few AFLP markers which were originally identified between clones AJ and AS [ 11 ] were not found in the progeny clones. Also, a few bands appeared in the progeny clones that were absent in the parents. All of these markers were ignored during the generation of the linkage map. It is possible that such markers could indicate genetic re-arrangements in the drug-resistant clones AS (3CQ) and AS (30CQ). However, they did not segregate with chloroquine resistance phenotype (data not shown). Some other markers were difficult to investigate because of their proximity to other bands or their location at the bottom of the gel, where bands tend to be fuzzier and more difficult to interpret. Effect of typing mistakes in the markers Ongoing work on linkage between chloroquine resistance and markers on chromosome 11 [ 4 ] suggested that AFLP and/or RFLP markers were occasionally incorrectly characterised in one ore more of the 28 clones. To test the effect of incorrect typing in AFLP markers, some deliberate mistakes were introduced by changing the parent from which a particular marker was inherited in a particular recombinant clone. The effect of such changes ranged from the appearance or disappearance of predicted double-crossover events and consequent change in the estimated genetic length, to a larger scale change in the order of markers within a linkage group. Occasionally markers were reallocated to a different linkage group. It was concluded that patterns of linkage may be sensitive to errors in genotyping individual clones. Discussion Genetic or physical linkage maps have been determined and reported for a number of apicomplexan parasites, including P. falciparum [ 20 ], P. c. chabaudi [ 11 ], Eimeria tenella [ 13 ], Toxoplasma gondii [ 23 ], Theileria parva [ 24 ] and Cryptosporidium parvum [ 25 ]. The map reported here is very extensive in terms of the numbers and density of markers included. Only the genetic map of P. falciparum exceeds its resolution Of the AFLP markers analysed, most were assigned to 10 of the 14 chromosomes, while some were placed on 12 unassigned linkage groups, which probably include groups located on chromosomes 2, 3, 4 and 14. The remaining AFLP markers could not be allocated to any linkage group. Unallocated AFLP markers and unassigned linkage groups may arise in a number of possible ways, discussed in the Results section above, including mistakes or considerable gaps in the recorded inheritance pattern of RFLP markers and, to a lesser extent, AFLP markers. Other factors include variations in the density of AFLP markers or polymorphism in general, areas of the genome where the rate of recombination is particularly high, linkage disequilibrium between loci on different chromosomes and non-random representation of clones in our sample. The numbers of markers allocated, the approximate genetic lengths and the number of recombination events were estimated for each of the linkage groups. The genetic length of the entire genome was estimated to be 1684 cM and the overall size of map unit 13.7 kb/cM. The genetic length and number of recombination events were expected to increase with the size of chromosomes. This was generally found to be the case, although a number of factors may influence these data. These factors include the failure to assign some linkage groups, incomplete or incorrect inheritance data, particularly for RFLPs, variation in the frequency of AFLP markers across the genome, variation in the recombination rate across the genome, and incorrect assignation of some linkage groups due to linkage disequilibrium. The presence and frequency of small indel mutations was confirmed. These markers could prove suitable for rapid typing of clones by size polymorphism and quantitative analysis by Real Time Quantitative PCR. The generation of a complete AFLP genetic linkage map for P. chabaudi was originally conceived as an essential step towards the identification of loci linked to genes encoding important phenotypes, such as drug resistance. Indeed, the identification of a locus underlying chloroquine resistance in P. chabaudi within approximately 250 kb of chromosome 11 [ 4 ] relied upon elements of the present map in the analysis of linkage between phenotype (chloroquine resistance) and genotype (inheritance of parental AFLP markers) in individual recombinant clones. However we have also developed a novel strategy called Linkage Group Selection [ 5 , 26 ] which more rapidly identifies loci linked to genes underlying selectable phenotypes, such as drug resistance. For example, a drug resistant parasite is crossed with a genetically different drug sensitive parasite. The uncloned recombinant progeny are drug treated, and AFLP markers which are linked to loci underlying drug resistance may be identified as those reduced in their representation or intensity [ 27 ]. A genetic linkage map enables us to determine whether AFLP markers which are significantly reduced in intensity lie in the same linkage group, prior to further sequence analysis. Genome sequence data are now available for P. c. chabaudi (partial) [ 8 ] and P. falciparum (complete) [ 7 ], and sequenced AFLP markers can sometimes be mapped to the P. falciparum genome. Because of the extensive gene synteny between P. chabaudi (and other rodent malarias) and P. falciparum [ 4 , 9 , 10 ], markers closely linked in P. falciparum are likely to be closely linked in P. chabaudi too. The correspondence between the genetic linkage map reported here and the mapping of AFLP markers to the P. falciparum genome in the studies discussed above [ 4 , 5 , 26 ] has increased our confidence both in the genetic linkage map reported here, and the extent of gene synteny between the P. c. chabaudi , Plsmodium yoelii and P. falciparum genomes [ 9 , 10 ]. The existence of a rodent malaria genome map, complete with syntenic relationships between it and the P. falciparum genome (Taco Kooij and Andy Waters, personal communication), will allow us to assign unallocated AFLP markers and unassigned linkage groups to particular chromosomes on the assumption that gene synteny is conserved. Authors' Contributions AM characterized the AFLP markers in the cross progeny, generated the genetic linkage map and drafted the article, PH helped in the generation of the linkage map, analysed genetic data from it and drafted the article, RF helped in the characterization of the AFLP markers in the cross progeny, PC and DW provided the recombinant clones and revised the article, RC designed and coordinated the study, revised the article and gave final approval. All authors read and approved the final manuscript. Supplementary Material Additional File 1 The file (.XLS) contains the following data for markers assigned to chromosomes and, where possible, for markers in unassigned linkage groups and the overall genome: – physical size of each chromosome (Mb), the number of markers (either RFLP or AFLP and total), the number of AFLP markers per Mb, the number of recombination events predicted for the 28 clones, the frequency of recombination events per Mb and per AFLP marker, the predicted genetic length of all linkage groups, and the estimated size of map unit (kb/cM). Click here for file Additional File 2 This file is the original PPT files from which figure 1 was derived.Figures 1-3 contain the linkage map for the chromosomes 1 and 5-13. Click here for file Additional File 3 This file is the original PPT files from which figure 2 was derived.Figures 1-3 contain the linkage map for the chromosomes 1 and 5-13. Click here for file Additional File 4 This file is the original PPT files from which figure 3 was derived.Figures 1-3 contain the linkage map for the chromosomes 1 and 5-13. Click here for file Additional File 5 This file is the original PPT files from which figure 4 was derived. Figure 4 contains various unassigned linkage groups. Click here for file	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC550669.xml
314476	Visualizing Noncentrosomal Microtubules during Spindle Assembly	null	As cells can only arise from cells that already exist, continuity of life depends on the highly regulated sequence of events that control cell division. This process is mediated by a complex macromolecular structure called the mitotic spindle. The most conspicuous components of the spindle are microtubules, which are made of tubulin and other associated proteins. In most animal cells—body cells and male germline cells (spermatocytes)—spindle assembly is orchestrated by organelles called centrosomes, which actively polymerize (that is, add tubulin subunits) and stabilize microtubules. The spindles found in these cells are known as astral because of the star-shaped asters—structures made of centrosome-anchored microtubules—that can be observed associating with each spindle pole. Some cells—such as the cells of the female germline (oocytes)—do not contain centrosomes, and the chromosomes themselves seem to arrange and stabilize the microtubules into spindles. These spindles are referred to as anastral . To gain insight into the mechanisms of spindle assembly, scientists are increasingly relying on techniques that allow them to directly observe dynamic, complex processes in the living cell. Using time-lapse microscopy of fluorescently labeled fruitfly ( Drosophila melanogaster ) spermatocytes, Cayetano Gonzalez and his colleagues at the European Molecular Biology Laboratory in Germany (and now at the Centro Nacional de Investigaciones Oncológicas in Spain) have been able to observe the assembly and sorting of microtubules of noncentrosomal origin in cells that contain centrosomes. The task of flagging such microtubules is complicated by the fact that centrosomes become quite active microtubule organizers once cell division begins. Thus, as soon as the membrane around the nucleus breaks down, microtubules from the centrosome invade the nuclear region, making it hard to identify any noncentrosomal microtubules that might appear. To get around this problem, Elena Rebollo in the Gonzalez lab set up two experimental conditions under which centrosomes remain functional but are kept affixed to the cell membrane—and, therefore, away from the nucleus—in Drosophila spermatocytes. One takes advantage of a genetic mutation (called asp , for abnormal spindle); the other uses a transient treatment with a drug (called colcemid) that depolymerizes microtubules. In these modified cells, microtubules can be seen growing not only over the membrane-bound centrosomes, as expected, but also over the nuclear region, away from the centrosomes. Nucleation, or formation, of such noncentrosomal microtubules has a relatively late onset, starting only once chromosomes are condensed, and takes place on the inner side of the remnants of the nuclear envelope. In a fraction of cells, these microtubules are sorted into bipolar spindle-shaped structures, highly reminiscent of the anastral spindles found in oocytes. Chromosome segregation—a critical stage of cell division—and cell division itself tend to be aberrant in these cells. These results, Rebollo et al. propose, strongly suggest that microtubules of noncentrosomal origin may significantly contribute to spindle assembly even in cells that contain active centrosomes. Moreover, by facilitating the nucleation of such noncentrosomal microtubules, the degraded nuclear envelope may play a previously unsuspected role in spindle assembly in Drosophila spermatocytes. It is unlikely, the researchers also conclude, that the anastral spindles they have observed can fill in as a backup to ensure successful cell division. More likely, they argue, both centrosomal and noncentrosomal microtubules are required for proper spindle assembly and robust cell division in cells with centrosomes. As the authors point out, Drosophila is a rich model system that should help scientists further investigate the intricacies of spindle assembly. The answers will help us understand how the cell executes one of its most important duties: safeguarding genomic stability for future generations. Centrosome-independent spindle assembly	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC314476.xml
538264	Gene expression profiling reveals novel TGFβ targets in adult lung fibroblasts	Background Transforming growth factor beta (TGFβ), a multifunctional cytokine, plays a crucial role in the accumulation of extracellular matrix components in lung fibrosis, where lung fibroblasts are considered to play a major role. Even though the effects of TGFβ on the gene expression of several proteins have been investigated in several lung fibroblast cell lines, the global pattern of response to this cytokine in adult lung fibroblasts is still unknown. Methods We used Affymetrix oligonucleotide microarrays U95v2, containing approximately 12,000 human genes, to study the transcriptional profile in response to a four hour treatment with TGFβ in control lung fibroblasts and in fibroblasts from patients with idiopathic and scleroderma-associated pulmonary fibrosis. A combination of the Affymetrix change algorithm (Microarray Suite 5) and of analysis of variance models was used to identify TGFβ-regulated genes. Additional criteria were an average up- or down- regulation of at least two fold. Results Exposure of fibroblasts to TGFβ had a profound impact on gene expression, resulting in regulation of 129 transcripts. We focused on genes not previously found to be regulated by TGFβ in lung fibroblasts or other cell types, including nuclear co-repressor 2, SMAD specific E3 ubiquitin protein ligase 2 ( SMURF2 ), bone morphogenetic protein 4 , and angiotensin II receptor type 1 ( AGTR1 ), and confirmed the microarray results by real time-PCR. Western Blotting confirmed induction at the protein level of AGTR1, the most highly induced gene in both control and fibrotic lung fibroblasts among genes encoding for signal transduction molecules. Upregulation of AGTR1 occurred through the MKK1/MKK2 signalling pathway. Immunohistochemical staining showed AGTR1 expression by lung fibroblasts in fibroblastic foci within biopsies of idiopathic pulmonary fibrosis. Conclusions This study identifies several novel TGFβ targets in lung fibroblasts, and confirms with independent methods the induction of angiotensin II receptor type 1, underlining a potential role for angiotensin II receptor 1 antagonism in the treatment of lung fibrosis.	Background Transforming Growth Factor beta (TGFβ) is a multifunctional cytokine that regulates a variety of physiological processes, including cell growth and differentiation, extracellular matrix production, embryonic development and wound healing [ 1 ]. Altered expression of TGFβ plays a crucial role in organ fibrosis, hypertrophic scarring, cancer, autoimmune and inflammatory diseases [ 2 ]. In the lung, TGFβ is consistently linked with progressive fibrosis [ 3 - 5 ]. Increased expression of TGFβ has been reported in a variety of fibrotic lung diseases [ 6 , 7 , 3 ], including idiopathic pulmonary fibrosis (IPF), a relentlessly progressive fibrotic lung disease with a median survival from diagnosis of only two years [ 8 ], and pulmonary fibrosis associated with systemic sclerosis, one of the leading causes of death in scleroderma patients [ 9 ]. Animal models also support a central role played by TGFβ in lung fibrosis. Intra-tracheal adenovirus-mediated TGFβ gene transfer causes severe lung fibrosis extending to the periphery of the lungs [ 5 ]. Mice lacking alphavbeta 6, an integrin which is crucial to the release of active TGFβ from latent extracellular complexes, develop lung inflammation but are strikingly protected from bleomycin-induced lung fibrosis [ 10 ]. IL-13 overexpression induces lung fibrosis which is mediated via TGF-β1 induction and activation [ 11 ]. Experimental inhibition of TGFβ with neutralizing antibodies, soluble receptors, or gene transfer of the TGFβ inhibitor Smad7, inhibits fibrosis in animal models [ 12 - 14 ]. Lung fibroblasts are the main cell type responsible for excessive extracellular matrix synthesis and deposition in fibrosing lung disorders [ 15 ]. TGFβ modulates fibroblast function through several mechanisms, including induction of extracellular matrix protein synthesis and inhibition of collagen degradation [ 1 ]. However, knowledge of TGFβ targets in adult lung fibroblasts is still limited to a small number of genes. Oligonucleotide array technology allows the simultaneous assessment of thousands of genes providing a global gene expression profiling of the response to a stimulus. The response to TGFβ has been investigated using oligonucleotide microarrays in keratinocytes [ 16 ] as well as in dermal [ 17 ] and in a human fetal lung fibroblast line [ 18 ], but not in primary human adult lung fibroblasts. Fibroblastic responses are likely to vary with the origin and developmental state of the cells [ 19 ], and a detailed study of TGFβ responses in adult lung fibroblasts is needed to gain further insights into the fibroproliferative process in the lung. We therefore quantified gene expression by oligonucleotide microarrays of adult lung fibroblasts (derived from biopsies of normal and both idiopathic and scleroderma-associated pulmonary fibrosis) in response to TGFβ, and identified several novel TGFβ targets among the wide variety of genes regulated by this cytokine. Of these, we particularly focused on angiotensin II receptor type 1 , the most highly TGFβ-induced gene among those encoding for signal transduction molecules. Methods Cell culture Primary adult lung fibroblasts were cultured from three control samples (unaffected lung from patients undergoing cancer-resection surgery) and from open-lung biopsy samples of lung fibrosis patients, three with idiopathic pulmonary fibrosis (IPF) [ 8 ] and three with pulmonary fibrosis associated with the fibrotic disease systemic sclerosis [ 9 ]. Independent reviews of the clinical (SV, ER) and histopathologic diagnosis (AGN) were performed. All the idiopathic pulmonary fibrosis biopsies were characterized by a usual interstitial pneumonia pattern (UIP), whereas all of the scleroderma-associated pulmonary fibrosis were classified as non-specific interstitial pneumonia (NSIP) [ 8 ]. Verbal and written consent was given by all subjects; authorization was given by the Royal Brompton Hospital Ethics Committee. Fibroblast culture conditions were as previously described [ 20 ]. At confluence, lung fibroblasts (all between passages 4–5) were serum-deprived for 16 hours, and exposed to either 4 ng/ml of activated TGF-β1 (R&D Systems) or serum-free culture medium for four hours. The concentration and time point of TGFβ used in our experiments was determined from ongoing studies within our laboratory, in which a 4 hour treatment with TGFβ 4 ng/ml was found to show significant induction of selected known direct TGFβ target genes, including CTGF. RNA isolation and gene array analysis At the end of the treatment period with or without TGFβ, total RNA was harvested (Trizol, Life Technologies), quantified, and integrity was verified by denaturing gel electrophoresis. Preparation of RNA samples for chip hybridization followed Affymetrix (Affymetrix, Santa Clara, California) protocols. Each RNA sample derived from an individual fibroblast line was hybridized on a separate microarray chip. Hybridization of cRNA to Affymetrix human U95Av2 chips, containing approximately 12,000 well characterized human genes, signal amplification and data collection were performed using an Affymetrix fluidics station and chip reader, following Affymetrix protocol. Scanned files were analyzed using Affymetrix Version 5.0 software (MAS5). Chip files were analyzed by scaling to an average intensity of 150 per gene, as recommended by Affymetrix. Reproducibility was assessed using two pairs of RNA samples from the same control line, TGFβ-treated/untreated; the concordance correlation coefficients were of 0.979 and 0.983, respectively. TGFβ response was analyzed by using a combination of the MAS5 Affymetrix change algorithm and of ANOVA models. According to Affymetrix criteria, in each TGFβ-treated/medium only pair, genes were defined as differentially regulated (either up or down) by TGFβ only when identified as significantly increased (I) or decreased (D) as determined by the Affymetrix change algorithm, with a change p value<0.001, and were detected as Present (according to the "absolute call"obtained by an Affymetrix algorithm) at least in the samples with the highest count (i.e. medium only in the case of D and TGFβ in the case of I). Genes were defined as TGFβ-responsive in normal human lung fibroblasts when they fulfilled all of the following three conditions: a) they were detected as TGFβ-regulated by Affymetrix criteria (see above) in at least two of the three control pairs; b) they showed a mean fold change after TGFβ of at least 2 (or lower than 0.5) in control fibroblasts; c) either a two-way ANOVA including only control fibroblasts detected a significant (p < 0.05) increase or decrease in control fibroblasts after TGFβ or they were also found to be responsive in at least four of the six fibrotic fibroblast lines and a significant effect (p < 0.05) of treatment (with TGFβ) was detected by a repeated measure ANOVA model including all the samples and adjusting for individual samples, disease, and interaction between treatment and disease. All statistical analyses were performed on log transformed data to reduce inequalities of variance. Thus, the latter ANOVA model could detect genes which were equally up- or down-regulated in normal and fibrotic fibroblasts, taking advantage of the larger number of samples, while the first model (equivalent to a paired t test) could detect changes possibly occurring in controls but not in fibrotic cell lines. Except for unknown genes, all gene symbols and names are given according to the nomenclature proposed by the Human Genome Organization (HUGO) Gene Nomenclature Committee. Real time-PCR Real time PCR (RT-PCR) was performed to confirm selected novel TGFβ targets in lung fibroblasts. Adult lung fibroblast lines [three control and three fibrotic (IPF)] were treated with or without TGFβ (4 ng/ml) for four hours. Total RNA was isolated from treated and untreated samples using Trizol (Life Technologies) and the integrity of the RNA was verified by gel electrophoresis. Total RNA (1 microgram) was reverse transcribed in a 20 μl reaction volume containing oligonucleotide dTs (dT 18 ) and random decamers (dN 10 ) using M-MLV reverse transcriptase (Promega) for 1 hour at 37°C. The cDNA was diluted to 100 μl with DEPC-treated water and 1 μl was used per real-time PCR reaction. A set of eight standards containing a known concentration of target amplicon was made by PCR amplification, isolation by gel electrophoresis through a 2% agarose gel followed by gel purification using QIAquick PCR purification spin columns (Qiagen). The concentration of the amplicon was measured by spectrophotometry and diluted in DEPC-treated water containing transfer RNA (10 μg/ml) to make standards of 10 fold dilutions from 100 pg/ μl to 0.01 fg/ μl. The target was measured in each sample and standard by real-time PCR using FastStart DNA Master SYBR Green (Roche Applied Science) as described by the manufacturer, in half the reaction volume (10 μl). Samples and standards were amplified for 30 to 40 cycles with the appropriate primers (Molecular Biology Unit, KCL School of Biological Sciences) at least in duplicate. The amount of target in the sample in picograms was read from the standard curve and values were normalised to 28S ribosomal RNA (pg of target/pg of 28S ribosomal-RNA). The oligonucleotide primer sequences are listed (5'-3'): angiotensin II receptor type1 ( AGTR1 ) primers: forward TGC TTC AGC CAG CGT CAG TT and reverse GGG ACT CAT AAT GGA AAG CAC; SMAD specific E3 ubiquitin protein ligase 2 ( SMURF2 ): forward AAC AAG AAC TAC GCA ATG GGG and reverse GTC CTC TGT TCA TAG CCT TCT G; nuclear receptor co-repressor 2 ( NCOR2 ): forward CAG CAG CGC ATC AAG TTC AT and reverse GTA ATA GAG GAC GCA CTC AGC; bone morphogenetic protein 4 ( BMP4 ) primers: forward CTA CTG GAC ACG AGA CTG GT and reverse GAG TCT GAT GGA GGT GAG TC. The results were analyzed using Student's paired t-test after logarithmic transformation, and statistical significance was taken as a p value of <0.05. Western blot analysis of TGFβ-induction of angiotensin II receptor 1 Lung fibroblasts were grown to confluence in DMEM with 10% FCS. At confluence, lung fibroblasts (all between passages 2–5) were serum-deprived overnight, and exposed to either 4 ng/ml of activated TGF-β1 (R&D Systems) or serum-free culture-medium with the addition of 0.1% BSA for 24 hours. To determine the signalling pathways through which TGFβ induces AGTR1, lung fibroblasts were treated with specific inhibitors 30 minutes before treatment with TGFβ. These included the dual MKK1/MKK2 inhibitor U0126 (10 μM) and predominant MKK1 inhibitor PD98059 (50 μM), known to inhibit MKK2 only weakly [ 21 ], as well as the p38 MAPK inhibitor SB 202190 (30 μM). Cell layer lysates were examined. Cell protein (10 μg/sample) was heated to 99°C for 5 min, loaded into sample wells, resolved on a 12% tricine SDS-polyacrylamide gel (Novex, San Diego, CA), and run at 120 V for 2 h. The separated proteins were transferred onto nitrocellulose membranes at 30V for 90 minutes. Membranes were blocked by incubation for one hour with 5% non-fat milk in phosphate buffered saline (PBS) containing 0.1% Tween 20. They were then washed and incubated overnight at 4°C in a 1:500 dilution of rabbit anti-angiotensin II receptor 1 polyclonal antibody (Santa Cruz Biotechnology), followed by a three-time wash in PBS and incubation in 1:1000 goat anti-rabbit biotinylated IgG (Vector Laboratories, Peterborough, UK) for 60 min at room temperature. Membranes were washed three times in PBS, and the signal was amplified/detected by using the ECL protocol as described by the manufacturer (Amersham plc, Little Chalfont, UK). Films were analysed by laser scanning densitometry on an Ultrascan XL (LKB-Wallac, UK). Data were analyzed by using Student's paired t test after log transformation and a p value<0.05 was considered significant. Immunohistochemistry The distribution of staining for AGTR1 was assessed by immunohistochemistry in surgical lung biopsies from four patients with idiopathic pulmonary fibrosis (IPF), meeting the diagnostic criteria of the American Thoracic Society/European Respiratory Society Consensus Classification [ 8 ], and in control biopsies (normal periphery of resected cancer) from three patients undergoing cancer resection surgery. Paraffin-embedded sections were dewaxed with xylene, hydrated and heated in the microwave at 120 degrees for 30 minutes in citrate buffer (10 mM pH 6.0). Slides were then briefly rinsed in PBS, blocked with 10% normal goat serum for 20', incubated with rabbit polyclonal anti-human AGTR1 antibody (N-10, 1:50, Santa Cruz Biotechnology, Santa Cruz, Calif) for one hour at room temperature. After washing with PBS, sections were incubated with biotinylated goat anti-rabbit IgG diluted in PBS (1:200) for 30 minutes, rinsed, and finally incubated with Vectastain Elite STR-ABC reagent (Vector Laboratories) for 30 minutes. After washing, sections were visualized using 3-amino-9-ethylcarbazole chromogen and H 2 O 2 as substrate (SK-4200; Vector Laboratories). Sections were then washed in tap water, counterstained with Carrazzis hematoxylin, and mounted with Gelmount (Biomeda, Foster City, CA) for examination using an Olympus BH-2 photomicroscope. Controls included an exchange of primary antibodies with goat matched antibodies. To confirm staining specificity, sections were also incubated with either nonimmune rabbit IgG control or secondary antibody only. Results Microarray analysis of TGFβ-response in primary adult lung fibroblasts According to the criteria outlined in the methods, a four hour treatment with TGFβ was found to regulate 129 transcripts in human lung fibroblasts. TGFβ-responsive transcripts included genes with roles in gene expression, matrix formation, cytoskeletal remodelling, signalling, cell proliferation, protein expression and degradation, cell adhesion and metabolism. A complete list of TGFβ-regulated genes is provided (see Additional file 1 ). The complete set of gene array data has been deposited in the Gene Expression Omnibus database with GEO serial accession number GSE1724 . We did not observe a substantial degree of difference in the response to TGFβ between the two fibrotic groups (idiopathic pulmonary fibrosis and scleroderma-associated pulmonary fibrosis) and control lung fibroblasts. Once the criteria outlined in the methods section and the p-value for interaction with treatment had been taken into account, there were no significant differences in the response to TGFβ among the three groups except for two genes, KIAA0261 (probe N: 40086_at), an unknown gene more upregulated in IPF (median fold change 2.2) than in scleroderma-associated pulmonary fibrosis (1.5) and in controls (1.3), and BTG1 (probe N: 37294_at), which was only slightly more downregulated in scleroderma-associated pulmonary fibrosis (fold change:0.4) than in IPF (0.6) and in controls (0.7). As both the number of genes and the magnitude of the differences were minimal, they were not considered meaningful and were not investigated further. Among genes responding significantly to TGFβ in control lung fibroblasts, as assessed by ANOVA analysis, none changed in opposite directions in either of the fibrotic groups. All the genes that responded significantly in the control group alone, were also TGFβ-responsive when analysis was extended to include the fibrotic cell lines. Furthermore, none of these genes responded differently to TGFβ between the two fibrotic groups, which are thus presented together in Tables 1 and 2 . Table 1 Transcription factor genes regulated by TGFβ in control and fibrotic lung fibroblasts (LF) Gene Symbol Affymetrix Probe N Control LF* Fibrotic LF* Gene name BHLHB2 40790_at 6.0 5.1 basic helix-loop-helix domain containing, class B, 2 CBFB 41175_at 2.9 2.8 core-binding factor, beta subunit EGR2 37863_at 52.0 3.3 early growth response 2 (Krox-20 homolog, Drosophila) ETV6 38491_at 2.0 2.6 ets variant gene 6 (TEL oncogene) FOXO1A 40570_at 3.8 6.0 forkhead box O1A (rhabdomyosarcoma) JUNB 2049_s_at 3.7 4.2 jun B proto-oncogene JUNB 32786_at 4.4 3.0 jun B proto-oncogene LRRFIP1 41320_s_at 2.1 1.5 leucine rich repeat (in FLII) interacting protein 1 MKL1 35629_at 2.7 2.6 megakaryoblastic leukemia (translocation) 1 MSC 35992_at 2.4 1.7 musculin (activated B-cell factor-1) NCOR2 39358_at 2.2 2.2 nuclear receptor co-repressor 2 NPAS2 39549_at 2.4 3.1 neuronal PAS domain protein 2 NR2F2 39397_at 0.4 0.5 nuclear receptor subfamily 2, group F, member 2 NRIP1 40088_at 2.3 1.8 nuclear receptor interacting protein 1 RUNX1 393_s_at 2.3 2.6 runt-related transcription factor 1 (aml1 oncogene) RUNX1 39421_at 3.1 2.3 runt-related transcription factor 1 (aml1 oncogene) RUNX1 943_at 2.2 2.7 runt-related transcription factor 1 (aml1 oncogene) SKI 41499_at 2.5 2.1 v-ski sarcoma viral oncogene homolog (avian) SMURF2 33354_at 2.2 2.2 E3 ubiquitin ligase SMURF2 SRF 1409_at 2.1 1.9 serum response factor SRF 40109_at 2.2 2.0 serum response factor TCF21 37247_at 0.2 0.4 transcription factor 21 TCF8 33439_at 2.8 1.8 transcription factor 8 (represses interleukin 2 expression) TIEG 224_at 2.2 2.1 TGFB inducible early growth response TIEG 38374_at 3.2 2.7 TGFB inducible early growth response ZFP36L2 32587_at 0.3 0.4 zinc finger protein 36, C3H type-like 2 ZFP36L2 32588_s_at 0.3 0.3 zinc finger protein 36, C3H type-like 2 ZNF365 35959_at 14.2 2.5 zinc finger protein 365 Mean fold change in mRNA abundance in TGFβ treated/untreated control and fibrotic lung fibroblasts (LF), respectively. Fibrotic lung fibroblast ratios represent the average values of idiopathic and scleroderma-associated pulmonary fibrosis lung fibroblasts. Table 2 TGFβ-regulated signalling and ECM/cytoskeletal genes in control and fibrotic lung fibroblasts Gene Symbol Affymetrix Probe N Control LF Fibrotic LF* Gene name Signal transduction ACVR1 39764_at 2.2 1.7 activin A receptor, type I ADM 34777_at 0.3 0.4 adrenomedullin AGTR1 346_s_at 3.8 3.2 angiotensin II receptor, type 1 AGTR1 37983_at 5.1 5.9 angiotensin II receptor, type 1 BDKRB2 39310_at 0.4 0.4 bradykinin receptor B2 BMP4 1114_at 0.2 0.2 bone morphogenetic protein 4 BMP4 40333_at 0.1 0.3 bone morphogenetic protein 4 DYRK2 40604_at 3.0 3.0 dual-specificity tyrosine-(Y)-phosphorylation regulated kinase 2 DYRK2 760_at 2.9 3.3 dual-specificity tyrosine-(Y)-phosphorylation regulated kinase 2 DYRK2 761_g_at 3.3 2.2 dual-specificity tyrosine-(Y)-phosphorylation regulated kinase 2 MLP 36174_at 2.4 1.7 MARCKS-like protein PLK2 41544_at 0.4 0.6 polo-like kinase 2 (Drosophila) RRAD 1776_at 3.0 5.2 Ras-related associated with diabetes RRAD 39528_at 3.6 5.1 Ras-related associated with diabetes SMAD3 38944_at 0.4 0.4 SMAD, mothers against DPP homolog 3 (Drosophila) SMAD7 1857_at 2.3 2.2 SMAD, mothers against DPP homolog 7 (Drosophila) SOCS1 41592_at 0.1 0.1 suppressor of cytokine signaling 1 SPRY2 33700_at 2.0 1.8 sprouty homolog 2 (Drosophila) STK38L 32182_at 3.7 3.8 serine/threonine kinase 38 like TGFBR3 1897_at 0.3 0.5 transforming growth factor, beta receptor III (betaglycan) TNFRSF1B 1583_at 0.4 0.6 tumor necrosis factor receptor superfamily, member 1B TNFRSF1B 33813_at 0.4 0.4 tumor necrosis factor receptor superfamily, member 1B TSPAN-2 35497_at 4.2 5.0 tetraspan 2 Extracellular matrix remodelling/Cytoskeletal COL4A1 39333_at 2.2 2.0 collagen, type IV, alpha 1 COMP 40161_at 2.7 5.3 cartilage oligomeric matrix protein COMP 40162_s_at 5.0 18.9 cartilage oligomeric matrix protein CTGF 36638_at 4.8 6.1 connective tissue growth factor CYR61 38772_at 4.4 3.5 cysteine-rich, angiogenic inducer, 61 ELN 31621_s_at 4.9 3.7 elastin ELN 39098_at 8.4 11.6 elastin PLAUR 189_s_at 2.7 2.8 plasminogen activator, urokinase receptor PLOD2 34795_at 2.5 1.8 procollagen-lysine, 2-oxoglutarate 5-dioxygenase 2 SERPINE1 38125_at 3.7 4.0 serine (or cysteine) proteinase inhibitor, clade E, member 1 SERPINE1 672_at 6.0 5.5 serine (or cysteine) proteinase inhibitor, clade E, member 1 TIMP3 1034_at 2.0 1.5 tissue inhibitor of metalloproteinase 3 TIMP3 1035_g_at 2.4 1.6 tissue inhibitor of metalloproteinase 3 TPM1 36790_at 2.3 1.7 tropomyosin 1 (alpha) TPM1 36791_g_at 2.7 2.1 tropomyosin 1 (alpha) TPM1 36792_at 2.5 2.0 tropomyosin 1 (alpha) Mean fold change in mRNA abundance in TGFβ treated/untreated control and fibrotic lung fibroblasts (LF), respectively. Fibrotic lung fibroblast ratios represent the average of idiopathic and scleroderma-associated pulmonary fibrosis lung fibroblasts. For the purpose of this study, we will concentrate on genes involved in transcriptional regulation, cytoskeletal/extracellular matrix organization, and signal transduction (Tables 1 and 2 ). Control of transcription TGFβ regulated a wide array of transcription factors (Table 1 ), including the known TGFβ target JUNB . Other TGFβ targets in lung fibroblasts identified by this study included Smad co-activators RUNX1 and CBFB , recently implicated in the targeted subnuclear localization of TGFβ-regulated Smads [ 22 , 23 ]. Transcriptional regulators involved in cell cycle control/cell differentiation induced by TGFβ included FOXO1A , NPAS2 , and TIEG ( TGFβ-inducible early growth response ), while ZFP36L2 , a zinc finger transcription factor linked to cell proliferation induction, was repressed by TGFβ. Serum response factor ( SRF ) and MKL1 were also induced by TGFβ. Transcriptional repressors induced by TGFβ included Ski , which together with Sno interacts with Smad molecules to inhibit transcription and may contribute to terminating TGFβ response [ 24 ] and TCF8 , a previously reported TGFβ target in fetal lung fibroblasts [ 18 ]. Other transcriptional co-repressors upregulated by TGFβ were nuclear co-repressors NCOR2 (or SMRT ) and BHLHB2 , which repress transcription by recruiting histone deacetylases [ 25 ], and musculin ( MSC ). Cytoskeletal/Extracellular matrix organization Most genes in this category were known TGFβ targets. As expected, transcripts involved in promoting extracellular matrix formation and cell adhesion such as connective tissue growth factor ( CTGF ) were upregulated, while we observed inhibition of bone morphogenetic protein 4 ( BMP4 ), a member of the TGFβ superfamily whose activity has recently been shown to be inhibited by CTGF through direct binding [ 26 ]. TGFβ also induced matrix genes including elastin ( ELN ), collagens ( COL4A1 ), plasminogen activator inhibitor ( PAI1 or SERPINE1 ) and PLOD2 , an enzyme which stabilizes collagen cross-links (Table 2 ). Tissue inhibitor of matrix metalloproteinase 3 ( TIMP3 ) was upregulated by TGFβ. Genes involved in cytoskeletal organization induced by TGFβ included known target tropomyosin ( TPM1 ). Interestingly, smoothelin , a smooth muscle gene recently reported to be highly induced by TGFβ in fetal lung fibroblasts [ 18 ], was also induced by TGFβ in this study, but at a slightly lower fold ratio than that chosen for the selection criteria (1.8). Control of signal transduction Among signalling molecules (Table 2 ), known targets included upregulation of SMAD7 and downregulation of SMAD3 [ 18 , 16 ]. Novel targets in lung fibroblasts included SMURF2 , a recently identified E3 ubiquitin ligase, which negatively regulates TGFβ signalling by targeting both TGFβ receptor-Smad7 complexes and Smad2 for ubiquitin-dependent degradation [ 27 , 28 ]. At the investigated timepoint, TGFβ downregulated the accessory receptor betaglycan , a membrane anchored proteoglycan which increases the affinity between TGFβ and type I and II receptors. Interestingly, TGFβ upregulated activin A type I receptor , a receptor for TGFβ family member activin, whose stimulation induces fibroblast-mediated collagen gel contraction [ 29 ]. Members of the Ras family of GTPases, ARHB and RADD (Ras-related GTP-binding protein), involved in cytoskeleton remodelling, were also upregulated by TGFβ. TGFβ also induced Dickkopf1 ( DKK1 ), a potent inhibitor of Wnt/beta-catenin signalling. Of particular interest was the novel observation that TGFβ upregulated angiotensin II receptor 1 ( AGTR1 ) in lung fibroblasts; conversely, the gene encoding for vasodilatory peptide adrenomedullin ( ADM ) was inhibited by TGFβ. Validation of selected TGFβ-induced genes by real time RT-PCR Several of the genes regulated by TGFβ confirmed previously published findings, thus validating our methods, including JUN-B , SMAD7 , connective tissue growth factor , elastin , and SERPINE1 [ 17 , 18 , 16 , 30 ]. To further consolidate our analysis, we selected a small group of novel TGFβ targets to be confirmed by RT-PCR in both control and fibrotic lung fibroblasts. These novel fibroblast TGFβ-responsive genes included potential key candidates in the regulation by TGFβ of lung tissue fibrosis and included angiotensin II receptor type 1 ( AGTR1 ), SMURF2 , a gene involved in terminating TGFβ signalling, NCOR2 , a transcriptional co-repressor and BMP4 , a member of the TGFβ family. Compared to untreated samples, we confirmed that TGFβ upregulated AGTR1 (ratio = 2.4; p = 0.002), SMURF2 , (ratio = 1.8, p = 0.003), NCOR2 (ratio 1.4; p = 0.004), and downregulated BMP4 (ratio = 0.4; p = 0.009), with no difference in the response between control and fibrotic fibroblasts (Figure 1 ). Figure 1 Independent verification of microarray results by measurement of gene expression with real time-PCR . TGFβ treatment (4 ng/ml) for four hours induces expression of mRNA for angiotensin receptor 1 (panel a), nuclear receptor co-repressor 2 ( NCOR2 ) (panel c) and SMURF2 (panel d) as well as inhibition of bone morphogenetic protein 4 (panel b) in three control lung fibroblast cell lines (dashed lines) and three fibrotic lung fibroblasts (solid lines). Induction of angiotensin II receptor type 1 by TGFβ We focused on AGTR1 protein because, as shown by microarray analysis, it was the most highly TGFβ-induced gene among signaling molecules in both control and fibrotic fibroblasts (Table 2 ). To verify whether AGTR1 mRNA upregulation corresponded to an increase in protein levels, we performed Western analysis on primary human adult lung fibroblasts exposed to TGFβ or medium alone in serum-free conditions for 24 hours. The intensity of the angiotensin II receptor 1 immunoreactive band was significantly increased in TGFβ-treated fibroblasts compared to those treated with medium alone (2.4 fold; p < 0.001) (Figure 2 ). To identify the signalling pathways through which TGFβ induces AGTR1, we evaluated whether the ability of TGFβ to induce AGTR1 expression in lung fibroblasts was blocked by specific signaling pathway inhibitors. A 30 minute preincubation with the dual MKK1/MKK2 inhibitor U0126 significantly inhibited TGFβ induction of AGTR1 protein (p < 0.01), whereas predominant MKK1 inhibitor PD98059 and p38 MAPK inhibitor SB202190 had no significant effect (Figure 2 ). Figure 2 TGFβ treatment induces angiotensin II receptor 1 (AGTR1) protein expression in adult lung fibroblasts; the induction is mediated by MKK1/MKK2 . Representative Western Blot (top) and average values (± SD) of angiotensin II receptor type 1 protein expression in lung fibroblasts treated with TGFβ (4 ng/ml)with or without 1/2 hour pre-incubation with of one the following signalling inhibitors: U0126, PD98059, SB202190. A 24 hour treatment with TGFβ induced an upregulation of AGTR1 protein (mean: 2.4 fold, *p < 0.001, Student's paired t-test). The induction of AGTR1 by TGFβ was specifically blocked by MKK1/MKK2 inhibitor U1026 (p < 0.01 compared with TGFβ-induced AGTR1, Student's paired t-test), but not by predominant MKK1 inhibitor PD98059 or p38 inhibitor SB202190). The results are representative of three independent experiments on both control and fibrotic cell lines. As a loading control, Western analysis with an anti-GAPDH antibody was also performed. AGTR1 expression in idiopathic pulmonary fibrosis lung biopsies We assessed staining for AGTR1 in lung biopsies from four patients with idiopathic pulmonary fibrosis and compared it to that of three control lungs. In particular we aimed to evaluate AGTR1 staining in fibroblastic foci, aggregates of fibroblasts/myofibroblasts in close contact with alveolar epithelial cells. Both in control and in idiopathic pulmonary fibrosis lung biopsies, AGTR1 immunoreactivity was observed in alveolar epithelial cells and alveolar macrophages. In addition, the fibroblasts within the fibroblastic foci present in idiopathic pulmonary fibrosis biopsies stained positive for the receptor (Figure 3 ). Figure 3 Angiotensin II receptor 1 staining in lung biopsies from control patients (A) and from patients with idiopathic pulmonary fibrosis (B) . Immunohistochemistry for the angiotensin II receptor 1 (AGTR1), counterstained with haematoxylin. AGTR1 positive staining is seen in alveolar macrophages, in epithelial cells and in fibroblastic foci (arrows) in usual interstitial pneumonia biopsies (panel B). Epithelial cells and alveolar macrophages express AGTR1 in control lung biopsies (panel A). Discussion In this study we report, for the first time, the transcriptional profile in response to TGFβ in adult primary human lung fibroblasts both from control and from fibrotic lungs. Our analysis of the response to TGFβ focused on TGFβ gene targets involved in transcription and signalling, identifying a series of genes previously unknown to respond to TGFβ in lung fibroblasts. These included angiotensin II receptor 1, providing further insights into links between TGFβ and angiotensin in the pathogenesis of fibrosis [ 31 , 32 ]. Although gene expression profiling in response to TGFβ has been investigated previously, earlier work has been confined to skin fibroblasts [ 17 ], keratinocytes [ 16 ], and a human fetal lung cell line [ 18 ], which is likely to respond differently to TGFβ from the adult lung fibroblast. Our data cannot be directly compared with the fetal lung fibroblast profiling because of methodological disparities, chiefly due to differences in the timing of the RNA collection. However, even restricting the comparison to results obtained at similar time points, we found a significant dissimilarity. Among transcription factors, only JUNB and TCF8 were upregulated by TGFβ both in fetal [ 18 ] and in adult lung fibroblasts, while all others differed between the two cell types. Interestingly, in this study, TGFβ caused an induction of both MKL1 and serum response factor , while neither were upregulated in fetal lung fibroblasts. The recently reported cooperation between these two transcription factors in determining smooth muscle cell differentiation [ 33 ] suggests that they may play a similar role in lung fibroblasts and suggests differences between fetal and adult lung fibroblasts in the transcriptional programs involved in the TGFβ-induced acquisition of the myofibroblastic phenotype. In this study, we did not observe a substantial difference in the response to TGFβ between lung fibroblasts from two patterns of fibrotic lung disease and control lung fibroblasts. In vivo heterogeneity between interstitial lung fibroblasts may occur in fibrotic and normal lung, obscuring the demarcation between normal and abnormal phenotypes, when cell lines are isolated using standard techniques [ 34 , 35 ]. This may explain discrepancies among studies on growth rate and resistance to apoptosis in fibroblasts derived from fibrotic lungs [ 34 , 36 ]. In particular, the fibroblasts/myofibroblasts forming the fibroblastic foci, observed to be linked to disease progression [ 37 ], could differ from the remaining fibroblasts found in the interstitium. The issue of sampling a population of homogeneous lung fibroblasts will be the subject of further investigation by using laser microdissection techniques targeting fibroblastic foci coupled with new technologies to amplify RNA from limited quantities of tissue [ 38 ]. Further, it is possible that the absence of striking differences in the response to TGFβ between disease groups and controls is due to a loss of the pro-fibrotic phenotype in vitro, even though the gene expression patterns of different passages of the same fibroblast line have been observed to cluster together, indicating that the in vitro phenotypes are stable through several passages in culture [ 19 ]. Further, we ensured that RNA was extracted from all fibroblast lines at comparable passages. Thus, even though our study cannot exclude the presence of subtle differences in the response to TGFβ, we have observed that, overall, fibrotic lung fibroblasts retain the capacity to respond to TGFβ, which could therefore be targeted by pharmacological means. Among the novel TGFβ targets identified by microarray analysis in lung fibroblasts, we focused our attention on the induction of angiotensin II receptor type 1 ( AGTR1 ), as its involvement is likely to significantly amplify the pro-fibrotic actions of TGFβ. The ligand for this receptor is angiotensin II, a vasoactive peptide which has been linked to fibrogenesis in the kidney and in the heart [ 39 , 40 ]. Recent studies have indicated that a local renin-angiotensin system could also be involved in the development of lung fibrosis [ 41 , 42 ]. Elevated angiotensin converting enzyme levels have been found in bronchoalveolar lavage (BAL) fluid from patients with idiopathic pulmonary fibrosis [ 41 ]. Compared to controls, lung fibroblasts from patients with idiopathic pulmonary fibrosis produce higher levels of angiotensin II, shown to induce apoptosis in alveolar epithelial cells through AGTR1 [ 31 , 43 ]. Blockade of angiotensin II or of AGTR1 attenuates lung collagen deposition in animal models of lung fibrosis [ 42 , 32 ]. Interestingly, the modulation of AGTR1 could be cell specific, as suggested by the report that TGFβ reduces AGTR1 expression in cardiac fibroblasts [ 44 ]. In addition to Smad molecules, the classic signalling pathway used by TGFβ family members, TGFβ also signals through the mitogen-activated protein kinase (MAPK) signalling pathways [ 16 ]. In this study, TGFβ was found to induce AGTR1 via mitogen-activated protein kinase kinase (MKK1/MKK2). The finding that the MKK1/MKK2 inhibitor U0126, but not the MKK1 inhibitor PD98059, was able to suppress TGFβ-induced AGTR1 expression, suggests that both MKK1 and MKK2 must be antagonized in order to inhibit transcription. The functional effects of AGTR1 stimulation in lung fibroblasts are only partially known. Although two isoforms of angiotensin II receptor exist, AGTR1 and AGTR2, the effects described so far of angiotensin II on lung fibroblasts are ascribed to the type 1 receptor. AGTR1 has been found to mediate mitogenesis in human lung fibroblasts [ 45 ] and extracellular matrix synthesis in lung [ 46 ] as well as in cardiac and dermal fibroblasts [ 47 ]. Whereas angiotensin II is known to induce TGFβ [ 46 ], the regulation of AGTR1 by TGFβ has not, to our knowledge, been previously reported in lung fibroblasts. Our data support the concept of a positive feed back loop by which TGFβ potentiates the pro-fibrotic actions of angiotensin II by increasing AGTR1 expression, providing a mechanism for the attenuation of the proliferative response to angiotensin II by TGFβ blockade [ 45 ]. Thus, cooperation and amplification of pro-fibrotic effects between TGFβ and AGTR1 are likely to be implicated in lung fibrosis. Interestingly, adrenomedullin, a multifunctional vasodilatory peptide that downregulates angiotensin II-induced collagen biosynthesis in cardiac fibroblasts [ 48 ], was inhibited by TGFβ, confirming a previous report [ 49 ], and suggesting that TGFβ exerts a complex regulation over vasoactive peptides and/or their receptors in lung fibroblasts. AGTR1 was found to localize to fibroblasts within fibroblastic foci in IPF/UIP biopsies. An increase in AGTR1 staining has been reported in the fibrotic regions surrounding the bronchioles in chronic obstructive pulmonary disease [ 50 ]. The finding that AGTR1 localizes to fibroblastic foci in IPF biopsies supports the potential relevance of the angiotensin system in this disease and suggests that the pro-fibrotic role of AGTR1 in IPF is not limited to epithelial cells [ 31 ]. Further studies are needed to assess the functional effects of AGTR1 stimulation in lung fibroblasts and to evaluate the biological role of AGTR1 in lung fibrosis. Conclusions Our findings confirm that in response to TGFβ, both control and fibrotic lung fibroblasts are potent effector cells expressing a very wide range of genes that are likely to contribute to the fibrotic process. In particular, we have shown that TGFβ has the capacity to influence the expression of angiotensin II receptor type 1 both at the mRNA and at the protein level. In view of the known induction of TGFβ by angiotensin II [ 45 ], our findings support the existence of a self-potentiating loop between TGFβ and angiotensin II, resulting in the amplification of the pro-fibrotic effects of both systems. Future treatment strategies could be based on the disruption of such interactions. Authors' contributions EAR participated in the design and interpretation of the study, carried out the cell culture work and participated in the microarray work, performed immunohistochemistry staining, and drafted the manuscript. DJA participated in the design and coordination of the study and in the preparation of the manuscript, SH performed the RT-PCR assays, XSW carried out the Western Blot analysis, PS performed the statistical analysis and participated in the interpretation of results and preparation of the manuscript, GBG participated in the microarray work, AUW participated in the interpretation of results, SV participated in cell line selection and clinical characterization, AGN reviewed fibrotic lung biopsies and interpreted immunohistochemistry staining, CD and CMB contributed towards the overall organizational setup for the study of lung fibroblast lines and participated in the interpretation of results, AL and JDP participated in the preparation of the manuscript, KIW conceived of the study and participated in the design, RdB participated in study design, interpretation and coordination. All authors read and approved the final manuscript. Supplementary Material Additional File 1 Complete list of genes regulated by a four hour treatment with TGFβ in control and fibrotic fibroblasts This data set contains all the genes up- or down-regulated by a four hour treatment with TGFβ (according to the criteria described in the methods) in control and fibrotic lung fibroblasts. Fibrotic lung fibroblast fold ratios are the average of the fold ratios for lung fibroblasts from idiopathic pulmonary fibrosis and pulmonary fibrosis associated with systemic sclerosis. Genes are sub-grouped into functional classes. Affymetrix probe set numbers, approved gene symbols, gene names and GenBank accession numbers are provided in the table. Click here for file	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC538264.xml
517495	Molecular clock in neutral protein evolution	Background A frequent observation in molecular evolution is that amino-acid substitution rates show an index of dispersion (that is, ratio of variance to mean) substantially larger than one. This observation has been termed the overdispersed molecular clock. On the basis of in silico protein-evolution experiments, Bastolla and coworkers recently proposed an explanation for this observation: Proteins drift in neutral space, and can temporarily get trapped in regions of substantially reduced neutrality. In these regions, substitution rates are suppressed, which results in an overall substitution process that is not Poissonian. However, the simulation method of Bastolla et al. is representative only for cases in which the product of mutation rate μ and population size N e is small. How the substitution process behaves when μN e is large is not known. Results Here, I study the behavior of the molecular clock in in silico protein evolution as a function of mutation rate and population size. I find that the index of dispersion decays with increasing μN e , and approaches 1 for large μN e . This observation can be explained with the selective pressure for mutational robustness, which is effective when μN e is large. This pressure keeps the population out of low-neutrality traps, and thus steadies the ticking of the molecular clock. Conclusions The molecular clock in neutral protein evolution can fall into two distinct regimes, a strongly overdispersed one for small μN e , and a mostly Poissonian one for large μN e . The former is relevant for the majority of organisms in the plant and animal kingdom, and the latter may be relevant for RNA viruses.	Background Kimura has argued that the majority of nucleotide substitutions that accumulate in genes over time are selectively neutral, and go to fixation purely by chance [ 1 ]. One major prediction of Kimura's neutral theory is that the substitution process should be a Poisson process, with the mean number of substitutions per unit time equal to the variance. In contrast to this theory, empirical studies often find the variance to be significantly larger than the mean [ 2 - 8 ]. This observation has been termed the "overdispersed molecular clock". (For an excellent review of both empirical evidence and mathematical theories, see Ref. [ 9 ].) It is possible to reconcile Kimura's theory with the overdispersed molecular clock via Takahata's fluctuating neutral space model [ 10 - 12 ]. If the neutral mutation rate fluctuates slowly, then the substitution process ceases to be Poissonian, and becomes indeed overdispersed. However, the problem with the fluctuating neutral space model is that it does not offer any argument for why the neutral mutation rate should fluctuate, and thus ultimately fails to explain the observed substitution patterns. An explanation for fluctuations in neutral mutation rate was recently proposed by Bastolla et al. [ 13 - 16 ]. Different proteins with identical structure naturally vary in their neutrality, that is, in the fraction of single-point mutants that are viable. Therefore, as a gene slowly drifts through sequence space, the neutral mutation rate will fluctuate in correlation to the changing neutrality, and this fluctuation alone could be sufficient to explain the overdispersed molecular clock. Bastolla et al. studied the substitution process in a variety of models of neutral protein evolution in silico , and found significant overdispersion in all cases they considered. However, the simulations that Bastolla et al. carried out were limited to cases in which the product of mutation rate μ and population size N e is small (because Bastolla et al. used only a single sequence as the representative of the whole population, an approach that is justified for μN e ≲ 1). Since population size and mutation rate can be substantial in some species (most notably in RNA viruses), it is justified to ask how general this result is for arbitrary values of μN e . It is known that large populations and populations evolving under high mutation pressure experience a strong selective pressure to avoid regions of low neutrality, an effect that has been termed "evolution of mutational robustness" [ 17 - 20 ]. In equilibrium, such populations settle in areas of sequence space that have above-average neutrality. As a result, regions of low neutrality are not represented in the population, and the distribution of neutralities in the population is much narrower than the total distribution of neutralities in sequence space. Therefore, we should expect that the neutral mutation rate does not fluctuate strongly under these conditions, and that the molecular clock will not be significantly overdispersed. For the present paper, I have studied the behavior of the substitution process under neutral protein evolution as a function of mutation rate μ and population size N e . I have found that the accumulation of non-synonymous mutations is substantially overdispersed for small μN e , in agreement with the results of Bastolla et al., but approaches a Poisson process when μN e ≫ 1. The accumulation of synonymous substitutions is always Poissonian, regardless of the value of μN e . Results I carried out simulations with DNA sequences of length L = 75. I determined the fitness of a DNA sequence by translating it into the corresponding amino-acid sequence, and determining its native fold within the framework of a lattice-protein model. (A sequence would have fitness 1 if it folded into a pre-determined target structure, and fitness 0 otherwise.) I used a simple model of maximally compact proteins on a 5 × 5 lattice. This protein-folding model is much simpler than the ones used by Bastolla et al. [ 13 , 14 ], but has been shown to produce realistic distributions of folding free energies and neutralities [ 21 - 23 ]. The advantage of the simpler model is that entire populations of evolving sequences can be simulated, instead of just individual sequences. First, I have found that my model produces overdispersion (that is, an index of dispersion R substantially above 1) for non-synonymous substitutions, but not for synonymous substitutions. The finding for synonymous mutations is not surprising, because changes in the protein's neutrality do not affect the probability with which a synonymous mutation is neutral (which is always one). Neutral evolution could produce overdispersion in the synonymous substitutions only if the number of synonymous sites in the sequence were undergoing significant fluctuations. While these fluctuations do occur, they are apparently not large enough to affect the index of dispersion. Second, I have found that for non-synonymous substitutions, R decays quickly with increasing population size N e at fixed μ (Fig. 1 ). Since one reason for a decaying index of dispersion could be a reduced number of accumulated mutations, I have studied how the mean number of accumulated mutations behaves as a function of population size. Instead of staying constant or decreasing, the mean increases with increasing N e , while the variance decreases (Fig. 2 ). This result shows that the reduction in R is not caused by a mere reduction in the accumulated mutations, and that the substitution process does indeed shift from overdispersed to Poissonian as the population size increases. Figure 1 Index of dispersion as a function of population size N e for synonymous and non-synonymous substitutions ( τ = 1000, μ = 0.075). Figure 2 Mean, and variance, of lineage-adjusted number of non-synonymous substitutions as a function of population size N e ( τ = 1000, μ = 0.075). Quantities were calculated from all 500 replicates at each population size. For non-synonymous substitutions, R decays with N e because of evolution of mutational robustness. However, mutational robustness is caused by large μN e , rather than large N e alone, and the parameter region in which mutational robustness becomes relevant is μN e ≳ 10 [ 17 ]. Therefore, it is more instructive to plot R as a function of μN e . The only problem with a naive plot of that sort is that R increases as a function of μτ , where τ is the length of the time window during which mutations accumulate [ 9 ]. Thus, in Fig. 3 , I show R for constant μτ as a function of μN e . Note that in this figure, instead of the sequence-wide mutation rate μ , I use the non-synonymous mutation rate μ n = 0.76 μ , which is corrected for the fact that only approximately 76% of mutations hit non-synonymous sites. (76% is the expected fraction of non-synonymous sites in a random DNA sequence.) Figure 3 shows that the transition from an overdispersed to a Poissonian substitution process occurs for μN e between approximately 10 and 100, in agreement with Ref. [ 17 ], and that the transition region seems to be largely independent of the value of μτ . Figure 3 Index of dispersion for non-synonymous mutations as a function of the product of non-synonymous mutation rate μ n (= 0.76 μ ) and population size N e . Figure 3 also shows that R increases with μτ . This dependency becomes clearer in Fig. 4 , where I display R as a function of μτ for fixed μN e . The figure shows substantial increase in R with increasing μτ for small to moderate μN e . However, even for μN e well above 50, there is still a slight increase in R with μτ . Therefore, my results do not settle the question of whether the substitution process becomes truly Poissonian for sufficiently large μN e , or whether it just approaches a Poisson process but always remains slightly overdispersed. To settle this question, one would have to carry out simulations with much larger τ and N e . Unfortunately, the protein folding model I use is still too computationally intensive to permit such simulations with current computational resources. Figure 4 Index of dispersion for non-synonymous mutations as a function of the product of non-synonymous mutation rate μ n (= 0.76 μ ) and divergence time τ . Discussion My results show that the size of the product μN e has a substantial effect on the index of dispersion under neutral evolution. The substitution process is strongly overdispersed for small μN e, but approaches a Poisson process as μN e grows large. Therefore, the next question is which of the two regimes has more biological relevance. As discussed by Cutler [ 9 ], the biggest problem in explaining the overdispersed molecular clock is not to come up with mechanisms that produce overdispersion, but to find a general mechanism that does not depend on special conditions or finely-tuned parameters. To assess the likelihood that fluctuations in neutrality contribute to the overdispersed molecular clock, we have to know the mutation rate and population size for the species of interest. It is notoriously difficult to obtain accurate data for these parameters, and only a few species have been studied in depth. One of the best data sets available is probably the one for Drosophila . Keightley and Eyre-Walker estimated the per-nucleotide substitution rate in Drosophila to be u = 2.2 × 10 -9 [ 24 ]. If we assume that the average gene in Drosophila is 1770 bp long [ 24 ], and that 76% of the nucleotides are non-synonymous (this number stems from averaging the number of non-synonymous sites over all codons with equal weight), then the average number of non-synonymous sites per gene is 1345 bp. Thus, the average rate of non-synonymous mutations per gene is μ n = 3.0 × 10 -6 . With an effective population size of approximately 3 × 10 5 [ 25 ], we get a product of population size and per-gene-non-synonymous mutation rate of approximately 1. Since selection for mutational robustness starts to take effect when this product is substantially larger than 1, Drosophila lies well within the parameter region in which we expect overdispersion to be caused by neutral evolution. For other higher organisms, in particular mammals, which tend to have comparatively small population sizes, we can expect that the product μ n N e falls into the same parameter region. On the other hand, for microorganisms, which can have very large population sizes, mutational robustness may play a role in their evolution. In particular, RNA viruses have genomic mutation rates on the order of one [ 26 , 27 ] and their genomes consist typically of only a handful of genes. Because RNA viruses undergo severe bottlenecks on a regular basis, their effective population size N e is much smaller than the number of virus particles in infected individuals (which can exceed 10 12 ), and is more closely related to the number of infected individuals. For HIV-1, N e has been estimated to be approximately 10 2 for subtype A, and 10 5 for subtype B [ 28 ]. The preceeding paragraph shows that neutral evolution of proteins is probably one source of overdispersed non-synonymous substitutions in Drosophila and other organisms. However, overdispersion has been observed in synonymous substitutions as well. For example, Zeng et al. [ 29 ] found an index of dispersion R significantly above one for synonymous, but not for non-synonymous substitutions in Drosophila . For mammals, some studies found R significantly above one for both synonymous and non-synonymous substitutions [ 8 ], while others found only the non-synonymous substitution process to be overdispersed [ 30 ]. Therefore, it is likely that other processes than neutral protein evolution also contribute to overdispersion. Such processes can be selection for optimal codon usage in the case of synonymous mutations, and positive selection on the amino acid level in the case of non-synonymous mutations. I have demonstrated that large μN e results in a substitution process with little overdispersion. However, I have not yet given an explanation for how overdispersion is reduced in populations with large μN e . There are two elements: First, selection for mutational robustness reduces the fraction of sequences with low neutrality, and increases the fraction of sequences with high neutrality, thus making the population more homogeneous and reducing the overall range of neutralities [ 17 - 20 ]. Second, a sequence with low neutrality will experience a real selective disadvantage in comparison to a sequence with high neutrality for large μN e , and will therefore have a reduced probability to end up on the line of descent. While this selective disadvantage is often small, it can nevertheless determine the evolutionary fate of a sequence in a large population. The larger the population, the more sensitive it becomes to small fitness differences, so that in a very large population a sequence with only a moderate reduction in neutrality will have a small probability to end up on the line of descent. (The fact that the mean substitution rate increases with the population size, as seen in Fig. 2 , is also consistent with this reasoning. The larger the population size, the more high-neutrality sequences end up on the line of descent, which is reflected in the increase in the mean substitution rate.) Throughout this paper, I have considered only neutral or lethal mutations. It is a reasonable question to ask if and how deleterious mutations would change my results. The answer is that they probably have only a minor impact, and the less so the larger N e , unless they are very slightly deleterious. In order to affect the molecular clock, the deleterious mutations must end up on the line of descent, that is, they must go to fixation. The probability of fixation p fix of deleterious mutants drops exponentially with the population size, p fix = [1 - exp(2 s )]/ [1 - exp(2 sN e )], where s is the selective disadvantage of the deleterious mutation [ 31 ]. Therefore, for reasonable population sizes, only very slightly deleterious mutations can go to fixation and thus affect the molecular clock. This reasoning is independent of the size of μN e , as long as N e is large in comparison to s . Conclusions The present study supports the following conclusions: • Neutral drift of proteins can lead to an overdispersed substitution process for non-synonymous mutations, but not for synonymous mutations. • The amount of overdispersion in the non-synonymous substitution process depends strongly on the product of mutation rate and population size. As this product increases, the substitution process becomes more and more Poissonian. The transition region starts at μN e ≈ 10, and extends to values well above 100. • It is not clear whether there are any species that have a sufficiently large population size and mutation rate to prevent overdispersion through neutral drift. In Drosophila , the product of mutation rate and population size is close to one, which is well below the parameter region in which the substitution process turns Poissonian. Methods Lattice protein model I implemented a version of the 5 × 5 lattice protein model put forward by Goldstein and coworkers [ 21 - 23 , 32 ]. In this model, proteins are sequences of n = 25 residues that fold into a maximally compact structure on a two-dimensional grid of 5 × 5 lattice points. There are 1081 distinct possible conformations in this model, and the partition function can be evaluated exactly by summing over the contact energies of all distinct conformations. The contact energy of a conformation i is where is the contact energy between amino acids at location j and at location k in the sequence, and is 1 if the two amino acids are in contact in conformation k , and 0 otherwise. The partition function is where the sum runs over all 1081 conformations. A sequence folds into conformation f if the contact energy for that conformation is lower than the contact energies of all other formations, E f < E i for all i ≠ f , and if the free energy of folding, which is defined as is smaller than some cutoff Δ G cut . Throughout this study, I used kT = 0.6 and Δ G cut = 0. The contact energies where taken from Table VI in Ref. [ 33 ]. Sequence evolution I simulated the evolution of populations of DNA sequences in discrete, non-overlapping generations. Population size is denoted by N e . The fitness of a sequence was 1 if the DNA sequence translated into a peptide sequence that could fold into a chosen target structure, and had a free energy of folding smaller than G cut . Otherwise, the fitness of the sequence was 0. All sequences had length L = 75. In each successive generation, sequences with fitness 1 were randomly chosen to reproduce, until the new generation had N e members. At reproduction, the sequences were mutated, with an average of μ base pair substitutions per sequence. I let each population evolve for several thousand generations, and kept track of the full genealogic information of all sequences in the population. In order to measure the molecular clock of fixed mutations only, I studied the pattern of base substitutions in a window of τ generations along the line of descent backwards in time, starting from the most recent common ancestor of the final population. I varied the parameters N e (10, 33, 100, 330, 1000, 3300), μ (0.0075, 0.075, 0.75), and τ (500, 1000). For each set of parameters, I carried out 500 replicates (each with a different, randomly chosen target structure), to obtain a distribution for the number of synonymous and non-synonymous substitutions S d and N d . Since there was some variation in the number of synonymous and non-synonymous sites across different target structures (on the order of approximately ± 5% variation from the mean), I then applied a correction factor to S d and N d to bring them into comparable units: I calculated the corrected number of synonymous substitutions as Here, S is the mean number of synonymous sites for the given replicate, and ( S ) is the average of S over all 500 replicates. Likewise, I calculated (Indices of dispersion calculated without this correction factor are slightly larger than the ones reported here, because the variation in S and N creates additional variance in S d and N d ). Similar correction factors have been used in sequence analysis [ 7 ], and are generally referred to as lineage adjustments. They control for differences among lineages that are primarily related to the expected number of substitutions in a lineage, and thus should not enter the index of dispersion. To obtain an estimate for mean and standard error of the index of dispersion, I subdivided the 500 results into 10 blocks of 50 each, and calculated mean and variance of the number of substitutions for each block. The ratio of variance to mean for a given set of substitutions (synonymous or non-synonymous) in a block is the index of dispersion for this data set. I then calculated mean and standard error for the index of dispersion from the individual results of the 10 blocks. The total CPU time needed to carry out all simulations was several months on a small cluster of Pentium II 500 MHz machines. Calculation of synonymous and non-synonymous substitutions and sites I calculated the number of synonymous and non-synonymous sites S and N and the number of synonymous and non-synonymous substitutions S d and N d according to the method proposed by Nei and Gojobori [ 34 ]. In short, under this method the number of synonymous sites s i of a codon i is the fraction of possible substitutions to that codon that leave the residue unchanged. The number of non-synonymous sites n i for the same codon is n i = 3 - s i . For the complete sequence, S and N are calculated as and where i runs over all codons in the sequence. The number of synonymous or non-synonymous substitutions s d, i or n d, i between two codons is the average number of such substitutions, where the average is taken over all paths that lead from one codon to the other. The total number of synonymous or non-synonymous substitutions between two sequences is the sum over all individual constributions, and (again, i runs over all codons in the sequence). To calculate the number of synonymous or non-synonymous substitutions along the line of descent, I simply summed up all synonymous or non-synonymous substitutions that occurred from generation to generation. Because the full evolutionary history was known, a correction for multiple mutations such as the Jukes-Cantor correction [ 35 ] was not necessary. I also averaged the number of synonymous and non-synonymous sites over all sequences along the line of descent, to get the mean number of synonymous and non-synonymous sites for the given evolutionary trajectory. Authors' contributions COW carried out all aspects of this study.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC517495.xml
550655	The PECACE domain: a new family of enzymes with potential peptidoglycan cleavage activity in Gram-positive bacteria	Background The metabolism of bacterial peptidoglycan is a dynamic process, synthases and cleavage enzymes are functionally coordinated. Lytic Transglycosylase enzymes (LT) are part of multienzyme complexes which regulate bacterial division and elongation. LTs are also involved in peptidoglycan turnover and in macromolecular transport systems. Despite their central importance, no LTs have been identified in the human pathogen Streptococcus pneumoniae . We report the identification of the first putative LT enzyme in S. pneumoniae and discuss its role in pneumococcal peptidoglycan metabolism. Results Homology searches of the pneumococcal genome allowed the identification of a new domain putatively involved in peptidoglycan cleavage (PECACE, PE ptidoglycan CA rbohydrate C leavage E nzyme). This sequence has been found exclusively in Gram-positive bacteria and gene clusters containing pecace are conserved among Streptococcal species. The PECACE domain is, in some instances, found in association with other domains known to catalyze peptidoglycan hydrolysis. Conclusions A new domain, PECACE, putatively involved in peptidoglycan hydrolysis has been identified in S. pneumoniae . The probable enzymatic activity deduced from the detailed analysis of the amino acid sequence suggests that the PECACE domain may proceed through a LT-type or goose lyzosyme-type cleavage mechanism. The PECACE function may differ largely from the other hydrolases already identified in the pneumococcus: LytA, LytB, LytC, CBPD and PcsB. The multimodular architecture of proteins containing the PECACE domain is another example of the many activities harbored by peptidoglycan hydrolases, which is probably required for the regulation of peptidoglycan metabolism. The release of new bacterial genomes sequences will probably add new members to the five groups identified so far in this work, and new groups could also emerge. Conversely, the functional characterization of the unknown domains mentioned in this work can now become easier, since bacterial peptidoglycan is proposed to be the substrate.	Background The bacterial cell wall resists intracellular pressure and gives the bacterium its particular shape. Cell wall reinforcement is brought about by a strong scaffolding structure, the peptidoglycan, which is formed by glycan strands and peptide chains held together by covalent bonds, resulting in a mono- or multilayered network. The glycan strands are composed of N -acetylglucosamine (Glc N Ac) and N -acetylmuramyl (Mur N Ac) residues linked together by β-1,4 glycosidic bonds. Peptides are covalently attached to the lactyl group of the muramic acid and their cross-linking results in the net structure of the peptidoglycan (Fig. 1a ). Figure 1 Schematic representation of peptidoglycan and of cleavage enzymes in S. pneumoniae . (a) Scheme of the pneumococcal peptidoglycan, indicating the chemical bonds cleaved by identified hydrolases in blue. The Mur N Ac residue containing the 1, 6-anhydro bond resulting from LT reaction is in a green circle. The putative LT pneumococcal enzyme appears in red, while enzymes CBPD and PcsB for which no enzymatic specificity is yet characterized are in black. (b) Topological representation of the glycan strand hydrolases described in S. pneumoniae . Black and hatched boxes indicate the signal peptide and the transmembrane anchor, respectively. The blue boxes illustrate the respective enzymatic active domains. Purple rectangles correspond to the Choline-Binding repeats. Green and orange boxes correspond to SH3b and coiled-coil regions, respectively. The topology was designed with the help of SMART server [39]. Peptidoglycan is synthesized in a multi-stage process. The first steps occur in the cytoplasm, where a set of enzymatic reactions gives rise to the assembly of the Mur N Ac-pentapeptide. This unit is in turn linked to the carrier undecaprenol lipid via a pyrophosphate group; afterwards the Glc N Ac group is added, generating the lipid II precursor. The saccharidic and peptidic moieties of lipid II are subsequently exposed to the periplasmic space. At this stage, peptidoglycan biosynthesis involves polymerization of the glycan chains, catalyzed by glycosyltransferases [ 1 ] as well as interpeptide bridge formation performed by transpeptidases [ 2 ]. These two enzymatic reactions are resident on the extracellular domains of Penicillin-Binding Proteins (PBPs) which are membrane-associated molecules, present in all eubacteria [ 2 ]. Peptidoglycan metabolism is a dynamic process since this structure grows and divides in perfect synchronization with cell growth and division. Furthermore, it is well established that peptidoglycan is subject to maturation, turnover and recycling in Gram-negative bacteria [ 3 ]. To fullfil these processes, it is expected that peptidoglycan cleavage enzymes must exert their functions in coordinated action with PBPs. Indeed, a large range of different peptidoglycan hydrolases have been identified in numerous bacterial species and specific peptidoglycan hydrolases exist for almost each covalent bond [ 3 ] (Fig. 1a ). The polysaccharidic component of peptidoglycan is the target of several hydrolases: the β-1,4 glycosidic bond between Mur N Ac and Glc N Ac residues is cleaved by lyzosyme and by lytic transglycosylases (LT), the β-1,4 glycosidic bond between Glc N Ac and Mur N Ac is hydrolyzed by glucosaminidases and amidases are responsible for the cleavage of the Mur N Ac-L-alanine bond (Fig. 1a ). Lyzosyme and LT enzymes cleave the same β-1,4-Mur N Ac-GlcNAc bond but generate different reaction products: while lyzosymes catalyze a hydrolytic reaction, LTs cleave the β-glycosidic linkage with the concomitant formation of 1,6-anhydromuramyl residues, blocking the reducing end of the glycan strands. The significance of the ring structure is not known but it has been speculated that the bond energy may be utilized for glycan strand rearrangements. In addition, the 1,6-anhydro ring may also be considered as a specific product of peptidoglycan turnover. Despite the lack of understanding of the physiological function of anhydromuropeptide product, LT enzymes must play a significant cellular role. Indeed, it has been observed that deletions of genes encoding LT proteins lead to E. coli and Neisseria meningitidis with altered cell separation phenotypes, indicating that LTs cleave septal peptidoglycan [ 4 , 5 ]. Macromolecular transport systems (secretion types II, III, IV and IV pilus synthesis) of Gram-negative bacteria contain LT enzymes, suggesting that peptidoglycan hole formation (essential for transport functions) is specifically performed by this enzyme family [ 6 ]. As mentioned above, the enlargement of the bacterial stress-bearing peptidoglycan structure requires the well coordinated action of synthases (PBPs) and hydrolase enzymes. The "three-for-one" growth mechanism described by Höltje proposes that a triplet of glycan strands cross-linked to each other (resulting from PBPs synthesis) is attached to the peptidoglycan layer. Subsequently, the docking strand is removed by hydrolases resulting in the insertion of the peptidoglycan triplet. The hydrolases involved in such multienzyme complexes are endopeptidases and LT enzymes [ 3 ]. This hypothesis is supported by experimental data as LT and PBPs could be co-purified from E. coli extracts [ 7 - 9 ]. In conclusion, LT enzymes play an important cellular role in diverse aspects of cell biology as expected from their presence in a very wide range of eubacteria as well as archaebacteria [ 3 , 10 , 11 ]. Surprinsingly, no such LT enzyme has been identified to date in the human pathogen Streptococcus pneumoniae , the causative agent of ear infections in children, as well as meningitis and pneumonia. The pattern of peptidoglycan hydrolases in this Gram-positive bacteria includes, besides a D, D-carboxypeptidase, five glycan strand cleaving enzymes (Fig 1b ). Four of these are surface-exposed proteins harboring Choline-Binding Domains which are non-covalently bound to choline residues present on cell wall pneumococcal teichoic and lipoteichoic acids [ 12 - 14 ]. The Choline-Binding Proteins (CBPs) catalyzing peptidoglycan hydrolysis are LytA, LytB, LytC and potentially CBPD (Fig 1b ). LytA is an amidase and also appears as an autolytic enzyme, causing bacteriolysis when acting in an uncontrolled manner [ 15 ]. LytB is a glucosaminidase involved in cell separation as lytB mutants form very long chains of over 100 cells [ 16 ]. LytC is a lysozyme with an autolytic behavior at 30°C [ 17 ]. Finally, CBPD and PcsB contain a CHAP domain (Cysteine, Histidine-dependent amidohydrolase/peptidase) predicted to hydrolyse the peptidoglycan in pneumococcus, but definitive biochemical data are still lacking [ 18 - 20 ]. Our interest in the biology of S. pneumoniae led us to investigate the presence of LT enzymes in this bacteria. Homology searches of enzyme sequences within the pneumococcus genome using bioinformatics tools allowed the identification of a new domain harboring motifs that infer potential peptidoglycan cleavage activity. For this reason we named this domain PECACE ( PE ptidoglycan CA rbohydrate C leavage E nzyme). This domain sequence was found exclusively in Gram-positive bacterial species, suggesting a significant cellular role. Finally, the PECACE domain is in some instances found in association with other domains, known to catalyze peptidoglycan hydrolysis: this observation reinforces the predicted function of PECACE as participating in peptidoglycan cleavage and represents another example of multifunctional proteins involved in peptidoglycan metabolism. Results and discussion Identification of a protein harboring the PECACE domain in S. pneumoniae The C-terminal domain of Escherichia coli Slt70 (Soluble Lytic Transglycosylase) has a lysozyme-like fold and its amino acid sequence was employed in a search of Bacilli genomes within the NCBI Conserved Domain Search server [ 11 , 21 - 23 ]. Thirty-four Slt70-homologue sequences were retrieved using an inclusion threshold of 0.01. None of these sequence originated from the S. pneumoniae translated genome. Subsequently, each of these 34 sequences was compared with the non-redundant protein database using PSI-BLAST with a E-value threshold of 0.005 and 5 sequences showed significant matches with a unique protein in S. pneumoniae . This sequence (accession numbers NP358524, gi:15902974) contains 204 amino acids: the first 21 amino acids are predicted to form a transmembrane anchor and the subsequent 192-residue region is putatively exposed to the extracellular space (Fig. 1b ). This S. pneumoniae NP358524 sequence has been tested as a pneumococcal vaccine antigen on the basis of preliminary screens for novel vaccine candidates [ 24 ]. A three-dimensional fold prediction of the S. pneumoniae NP358524 protein was performed with the 3D-PSSM server [ 25 ] which identified two matches: E. coli Slt70 (d1qsaa2, E-value:10 -7 ) and LysG (G-type goose lyzozyme, d1531, E-value:10 -3 ). The sequence alignment between NP358524 and Slt70 is shown in Fig. 2 , defining the PECACE domain in the pneumococcal protein. The secondary structures are also reported, based on three-dimensional structures of Slt70 and on computational predictions for PECACE and suggest that the latter is highly α-helical (Fig. 2 ). It is of note that both Slt70 and LysG are highly similar, and both lack the catalytic aspartate residue commonly found in the active site of lysozymes [ 10 , 11 , 21 , 22 ]. Therefore, the PECACE domain of the NP358524 sequence appears to belong to this group of bacterial lysozymes, characterized by the absence of an aspartate residue in the catalytic site and is part of the Glycoside Hydrolase family 23 based upon CAZy classification . The catalytic acid residue in the PECACE domain is most probably Glu61 since it aligns with the catalytic Glu478 residue in the Slt70 sequence (Fig. 2 ). The serine residue following the catalytic glutamate and the GLMQI/V motif are essential for active-site architecture and are conserved between Slt70 and LysG. In the PECACE sequence, a threonine residue follows the catalytic glutamate and the GLMQI/V motif differs since the corresponding sequence is D(68)VMQS (Fig. 2 ). Finally, the second motif AYNxG which has been shown to be involved in the interaction with the substrate for Slt70 (A551YNxG) is well conserved in the PECACE sequence (A117YNxG). Figure 2 Alignment of the PECACE domain with Slt70. Protein fold recognition was performed with the 3D-PSSM server. The NP358524 sequence (residues 31–145) from S. pneumoniae (PECACE domain) is aligned with Slt70 from E. coli (P03810, residues 478–616). Amino acids of Slt70 involved in the catalytic reaction and in ligand recognition are underlined while residues conserved in each alignment are highlighted in red. The structural prediction for S. pneumoniae PECACE domain was determined (H = helix, C = coil) while Slt70 secondary-structure information was obtained from PDB file 1QSA. Based on this sequence analysis, we infer that the S. pneumoniae NP358524 protein, through its PECACE domain, probably catalyzes the peptidoglycan cleavage of the β-1,4-Mur N Ac-Glc N Ac bond by employing Glu61 as the catalytic residue. Identification of the PECACE domain in Gram-positive bacteria The 204 amino acid sequence from S. pneumoniae NP358524, containing the PECACE domain, was used as a PSI-BLAST search query. In total, 29 distinct proteins, all from Gram-positive bacteria, were identified (E-value: 10 -5 ) and no sequences from Gram-negative bacteria were retrieved. These sequences were aligned with ClustalW and manually edited. A conserved pattern could be extracted from this alignment: E- [ST]-X-G-X(1,16)-D-X-M-Q- [SA]- [SA]-E- [SG] which was used to search for additional sequences, but no new sequence could be detected from databases, even with a degenerated pattern. PSI-BLAST performed through the GOLD server led to the identification of 10 new sequences from Gram-positive bacteria [ 26 ]. In summary, out of the about 50 Gram-positive bacteria for which the whole genome sequence is available, 34 of them contain at least one protein harboring the PECACE domain. The final alignment of these sequences with the S. pneumoniae PECACE domain is shown in Fig. 3 . The putative catalytic glutamate residue, Glu61 in the S. pneumoniae PECACE domain, is conserved in all sequences and the following residue is a Ser or Thr in accordance with Slt70 and LysG patterns. In addition, the D(68)VMQS motif in the S. pneumoniae PECACE domain is also well represented in the large majority of sequences with the consensus sequence DI/VMQSSES. Finally, the second motif AYNxG is also conserved while the Ala residue is often replaced by a Ser. In conclusion, the features identified in the S. pneumoniae PECACE domain regarding the potential enzymatic properties of peptidoglycan polysaccharide cleavage are also shared by the similar PECACE domains in Gram-positive bacteria. Figure 3 Sequence alignment of PECACE domains identified in Gram-positive bacteria . Multiple sequence alignment was constructed using ClustalW. The lengths of the insertions in the sequences are shown in parentheses. The sequences are denoted by their GenBank Identifier (gi). The domain limits are indicated by the residue positions (first-end). The amino acids identified as catalytic or involved in ligand recognition are marked with asterisks under PECACE sequence. Alignments are coloured using the CHROMA tool using default parameters [40]. Full sequence details, group (i): Streptococcus pneumoniae R6 (gi:15902974), Streptococcus mitis NCTC 12261 (§SMT1418), Streptococcus sanguinis SK36 (&:SS_A352_G10), Streptococcus gordonii (gi:18389219), Streptococcus suis P1/7 (suis166b12), Streptococcus uberis 0140J (sub49a04), Streptococcus equi (equi324d3), Streptococcus equi subsp. Zooepidemicus (zoo26g07), Streptococcus pyogenes M1 GAS (gi:15675124), Streptococcus agalactiae 2603V/R (gi:22537230), Lactococcus lactis subsp. Cremoris SK11 (scaffold18), Streptococcus mutans UA159 (gi:24379517), Streptococcus thermophilus LMD-9 (scaffold3), Lactococcus lactis subsp. Lactis (gi:15672584), Enterococcus faecium DO (2351355_Cont543), Enterococcus faecalis V583 (gi:29376084), Bacillus subtilis subsp. subtilis str. 168 (gi:16078973), Bacillus cereus ATCC 14579 (gi:30020591), Oceanobacillus iheyensis HTE831 (gi:23100516), group (ii): Bacillus anthracis : (pXO2-08) (gi:10956398), Enterococcus faecalis : (pRE25) (gi:12957015), Enterococcus faecium (gi:22992993), Enterococcus faecalis V583 (gi:29376781), Clostridium difficile 630 (Cd81d2), Enterococcus faecalis V583 (gi:29376405), Clostridium perfringens (gi:13274506), Staphylococcus aureus subsp. aureus Mu50 (gi:15923390), Listeria monocytogenes EGD-e (gi:16803144), Streptococcus agalactiae 2603V/R (gi:22537089), Enterococcus faecium (gi:22993467), Bacillus subtilis subsp. subtilis str. 168 (gi:16077564, group (iii): Bacillus cereus ATCC 14579 (gi:30021796), group (iv): Enterococcus faecalis BM4518 (gi:33355845), group (v): Bacillus anthracis str. A2012: (pXO1) (gi:21392795), Bacillus cereus ATCC 10987 : (pBc10987) (gi:44004362). Genomic organization of pecace genes The genomic organization of pecace genes has been analyzed in a variety of Gram-positive bacteria and a conserved distribution was observed in various streptococci species (Fig. 4 ). This feature indicates that genetic transfer of the whole cluster may have occured within the streptococci family, providing further evidence regarding the significant importance of the PECACE domains in bacterial physiology. However, the pneumococcal cluster is more related to the S. mitis one than to S. mutans , S. agalactiae and S. pyogenes ones, while clusters of the latter three species are related to each other. Genes located upstream and downstream of pecace are in some instances well characterized but the function of the corresponding proteins could not bring any clues about the role of PECACE, nor any evidence on pecace gene transcription. However, pecace is in all cases found in association with the same gene (whose locus name in S. pneumoniae is spr0929) but no information about the function of the protein encoded by this locus is available in databases. Transcriptional analysis of these two genes may bring informations about their potential co-regulation, a first stage in deciphering cellular function. Figure 4 Schematic representation of the gene cluster containing pecace in Streptococci species. The coding regions and their direction of transcription are indicated by arrows. Gene names are given on top of the corresponding region. Domain organization of proteins containing the PECACE domain The PECACE domain is found in a large range of protein architectures, commonly associated with other peptidoglycan hydrolases, suggesting that these proteins have multiple peptidoglycan cleavages activities (Fig. 5 ). The identification of proteins displaying the PECACE domain was carried out using NCBI Conserved Domain Search and Pfam servers. In addition, prediction of membrane anchoring was performed with the DAS-Transmembrane Prediction server while extracellular secretion of the protein was deduced from the identification of a signal peptide. Figure 5 Domain architecture of PECACE proteins. The domain architecture of the proteins containing the PECACE domain was organized according to searches with NCBI Conserved Domain Search server against Pfam database: CHAP/NlpC-P60 (Pfam: PF05257/PF00877), M37 peptidase (Pfam: PF01551), unknown domain 1 (gi: 33355845) and unknown domain 2 (gi: 30021796). The size of the domains is not respected in these representations. PECACE-containing proteins appear to fall into 5 main categories (Fig. 5 ): (i) those which display no additional domain, (ii) CHAP-Nlpc/P60 as the associated group, (iii) CHAP-Nlpc/P60 and an unknown domain as associated groups, (iv) domains with no ascribed functions and finally (v) CHAP-Nlpc/P60 and M37 peptidase as associated groups. The 19 proteins which contain only the PECACE domain belong to group (i) and harbor either a signal peptide or a transmembrane helix (as for the S. pneumoniae protein), leading in both cases to cell surface expression. The CHAP-Nlpc/P60 domain is commonly associated with the PECACE domain in different modular organizations, namely in groups (ii), (iii), and (v) [ 18 , 19 , 27 ]. The CHAP domain has been recently described as a Cysteine, Histidine-dependent Amidohydrolase/Peptidase and it has been proposed to hydrolyse peptidoglycan containing γ-glutamyl [ 18 , 19 ]. Indeed, proteins such as N -acetylmuramyl-L-alanine amidase and D-alanyl-glycyl endopeptidase have been described as CHAP-containing enzymes [ 18 , 19 ]. However, while the substrate and the reaction mechanism have not been yet experimentally characterized for the CHAP domain, its role in peptidoglycan hydrolysis is inferred from its presence in multifunctional proteins recognizing peptidoglycan as substrate. Recently, hydrolytic activity of peptidoglycan has been attributed to the CHAP-containing protein PcsB in S. pneumoniae due to abnormal and uncontrolled cell wall synthesis at misplaced septa and formation of long cells in pcsB deleted mutant strains [ 20 ]. Proteins from group (ii) are expressed at the cell surface through a transmembrane anchor or are secreted, 12 members have been identified with this topology. Only one sequence (AAQ16265, gi:33355845) from Enterococcus faecalis BM4518 is part of the group (iii), and no function could be identified for the N-terminus domain preceding the PECACE domain. However the former domain is Lys-rich (14%) suggesting an electrostatic interaction with the peptidoglycan as proposed for B. subtilis endopeptidase [ 28 ]. Group (iv) is composed of an unique sequence from B. cereus ATCC 14579 (NP 833427, gi:30021796). Neither a signal peptide nor a transmembrane anchor have been detected. Furthermore, the domain of unknown function, which is different from the ones identified in groups (iii) and (v) is present in other multimodular proteins of B. cereus , in association with peptidoglycan hydrolysis enzymes. Finally, two sequences share the architecture defining the group (v) which harbor CHAP-Nlpc/P60 and Peptidase M37 domains [ 29 ]. Members of the Peptidase M37 family are generally glycylglycine endo-metallopeptidases; the archetypal member is the lysostaphin enzyme from Staphylococcus species which cleaves the pentaglycine bridge in the peptidoglycan [ 30 ]. One group (v) protein (NP 652875, gi:21392795) is encoded by Bacillus anthracis plasmid pXO1 and is required for synthesis of various anthrax toxin proteins [ 31 ]; this sequence has neither a signal peptide nor a transmembrane region. The second sequence of group (v) is located on Bacillus cereus ATCC 10987 plasmid pBc10987 (NP 982030, gi:44004362) and contains, in addition to CHAP-Nlpc/P60, Peptidase M37 and PECACE domain as well as an extra sequence to which no function has been attributed but with significant similarity with a B. anthracis plasmid pXO1 sequence (NP 652874, gi:21392794) [ 32 ]. Conclusions In summary, a new domain named PECACE, putatively involved in peptidoglycan cleavage has been identified in S. pneumoniae . The probable enzymatic activity deduced from the detailed analysis of the amino acid sequence suggests a LT-type or goose lyzosyme-type mechanism; we are currently characterising the enzymatic properties and cellular role of the PECACE domain from S. pneumoniae . This new putative pneumococcal peptidoglycan cleavage enzyme differs largely from the other hydrolases already identified in this bacteria. Indeed, LytA, LytB, LytC and CBPD proteins are all bound to the cell wall choline residues and thus expressed at the cell surface. The presence of a signal peptide within the amino acid sequence of PcsB suggests that it is either exposed on the cell surface or secreted. On the contrary, the pneumococcal NP358524 protein displaying the PECACE domain is embeded in the cytoplasmic membrane by a hydrophobic helix. The physiological role of this membranous peptidoglycan cleavage enzyme might differ from the other peptidoglycan hydrolysing enzymes. Interestingly, the PECACE domain has only been found in Gram-positive bacteria. It is tempting to speculate that the multilayered structure of Gram-positive peptidoglycan relates to the PECACE putative activity. The architecture of multimodular proteins containing the PECACE domain is another example of the pattern of multiple activities harbored by many peptidoglycan hydrolases, probably needed for the regulation of peptidoglycan metabolism. The release of new bacterial genomes sequences will probably add new members that will complete the five groups identified so far in this work and new groups could also emerge. Conversely, the functional characterization of the unknown domains mentioned in this work should now be easier, as their substrate, the peptidoglycan, is now identified. Methods The non-redudant database of protein sequences (National center for Biotechnology Information, NIH, Bethesda) and whole bacterial genomes sequences [ 26 ] was searched using BLASP and PSI-BLAST programs with (E) value threshold of 0.005 [ 33 ]. Multiple alignments were constructed with ClustalW program [ 34 ] followed by manual correction based on PSI-BLAST results. Protein fold recognition through 3D-profiles was searched using 3D-PSSM server [ 25 ]. Conserved (and degenerated) amino acid patterns was designed and searched against non-redudant database of protein sequences . Identification of domains associated with PECACE proteins was realized using NCBI Conserved Domain Search [ 35 ] and Pfam servers [ 36 ]. Finally, prediction of transmembrane anchor and secretory signal peptide were performed with DAS server and SignalP-2.0 servers respectively [ 37 , 38 ]. Authors' contributions OD conceived of the study and participated in the sequences alignment. EP carried out the sequences analysis and the writing of the manuscript with AMDG. TV coordinated the study. All authors read and approved the final manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC550655.xml
503386	Attentional influences on functional mapping of speech sounds in human auditory cortex	Background The speech signal contains both information about phonological features such as place of articulation and non-phonological features such as speaker identity. These are different aspects of the 'what'-processing stream (speaker vs. speech content), and here we show that they can be further segregated as they may occur in parallel but within different neural substrates. Subjects listened to two different vowels, each spoken by two different speakers. During one block, they were asked to identify a given vowel irrespectively of the speaker (phonological categorization), while during the other block the speaker had to be identified irrespectively of the vowel (speaker categorization). Auditory evoked fields were recorded using 148-channel magnetoencephalography (MEG), and magnetic source imaging was obtained for 17 subjects. Results During phonological categorization, a vowel-dependent difference of N100m source location perpendicular to the main tonotopic gradient replicated previous findings. In speaker categorization, the relative mapping of vowels remained unchanged but sources were shifted towards more posterior and more superior locations. Conclusions These results imply that the N100m reflects the extraction of abstract invariants from the speech signal. This part of the processing is accomplished in auditory areas anterior to AI, which are part of the auditory 'what' system. This network seems to include spatially separable modules for identifying the phonological information and for associating it with a particular speaker that are activated in synchrony but within different regions, suggesting that the 'what' processing can be more adequately modeled by a stream of parallel stages. The relative activation of the parallel processing stages can be modulated by attentional or task demands.	Background This study explores attentional modulation within the 'what'-stream of the auditory modality during phoneme processing. Knowledge of speech sound representation in the auditory domain is still sparse. However, parallels to the extensively studied visual modality and also to the somatosensory domain are becoming evident. For example, columnar mapping of several stimulus properties (as known from the visual cortex) has been revealed in human and animal research: acoustic parameters like spectral bandwidth, periodicity, stimulus intensity [ 1 , 2 ] or – for human speech sounds – distance between spectral peaks [ 3 , 4 ] appear to be mapped perpendicularly to the main cochleotopic gradient. Recently, a segregation of a ventral 'what' and a dorsal 'where' stream – as long established in the visual system [ 5 ] – has also been proposed for the auditory system. This conclusion was based on neuroanatomical and functional studies in macaques [ 6 - 8 ] and has been substantiated in humans [ 9 , 10 ]. Given these parallels between sensory domains and the increasing preference for complex stimuli along the auditory central pathway, more complex topologies such as language-specific maps in auditory cortex are also plausible, and evidence for individually ordered mapping of speech sounds is growing [ 11 - 15 ] (for speech-specific vocalizations in animals see [ 8 , 16 ]). More specifically, data from our lab imply map dimensions along phonological features which build the basic components of speech sounds: In Obleser et al. [ 15 ], responses to DORSAL vowels (which are articulated with the back of the tongue and which exhibit a small distance between spectral peaks, i.e., small F 1 -F 2 distance) were located more posterior in auditory association cortex than responses to CORONAL vowels (which are articulated with the tip of the tongue and which exhibit a large distance between spectral peaks, i.e., larger F 1 -F 2 distance), and a topographical shift between these classes of vowels even when embedded in non-words has been reported [ 15 , 17 ]. Research has long been tackling the question of attention and attentional top-down modulation that may tune cortical neurons and with it functional maps in a context-specific manner: In the visual domain, a top-down influence on receptive fields of areas as basic as VI has been shown [ 18 , 19 ], and in the somatosensory domain Ergenzinger and colleagues reported that drastic changes in functional maps can be experimentally induced even on a thalamic level [ 20 ]. The thalamic homuncular representation of a monkey's hand becomes blurred and distorted when top-down modulation from somatosensory cortex is blocked neurochemically within the cortex. These results emphasize the possibility of attention-dependent modulation of maps, a topic exemplified in a somatosensory MEG mapping study by Braun and colleagues [ 21 ]: In a somatosensory stimulation with small brushes moving back and forth across the digit tips, subjects either attended the movement of single brushes on single digits and reported the movement direction or they attended and reported the global direction of all brushes on all five digits. Magnetic source imaging of the somatosensory evoked field revealed a typical homuncular representation of the single digits spread along the post central gyrus only in the condition where the focus of attention was on single digits rather than on the hand as a whole. In the latter condition, top-down attentional demands temporarily seemed to blur the single digit mapping. For the developing field of speech sound mapping, top-down influences of attentional demands on functional organization at the different stages in the processing streams have not been sufficiently studied. Nevertheless, it becomes a central issue when the functional architecture of the effortless and robust perception of speech shall be understood. It is common to study speech perception either in passive oddball paradigms [ 22 , 23 ] where the subject's attention is deliberately forced to a movie or to reading a book, or in passive listening conditions where no attentional control is experimentally induced (e.g. [ 24 , 25 ]), or in active target detection tasks where the attention is commonly focused on the phonological content of the speech material [ 14 , 15 , 26 ]. We analyzed the magnetic N100 (N100m) response to two vowels [o] and [ø], both produced by a male and a female speaker. Subject's attention was either on the vowel or on the speaker difference, in a counterbalanced order. How would a controlled shift of attention from specific phonological features of speech to features of speaker identity affect the speech sound mapping in timing and topography of the brain response? Two concurrent outcomes are conceivable here: First, from the numerous parallels between the auditory and other sensory domains, one might expect a blurring of differences of the phonological map in auditory cortex when features such as the speaker identity rather than phonological differences are attended over minutes. Second, phonological processing could be the default process needed in all speech-listening situations and should therefore activate phonological feature maps irrespectively of attentional demands. We would then expect that the separate mapping of DORSAL and CORONAL vowels described previously [ 15 ] is unaffected by an attentional focus on speaker identity. However, a shift of activational patterns as an entity would reveal more about the staging of parallel processing in the flow of the 'what' stream. Results In 21 of 22 subjects, a clear waveform deflection around 100 ms post vowel onset was observed (Fig. 2 ) in all conditions over both hemispheres and sensor space parameters peak latency and amplitude were obtained. Satisfying and physiologically plausible dipole fits (see methods) in both hemispheres could be obtained in 17 subjects and were subjected to statistical analysis. N100m latency, amplitude and source strength Analysis of the N100m root mean square (RMS) peak latency revealed foremost a main effect of vowel (F 1,20 = 44.8, p < .0001, Fig. 2 ), whereby the DORSAL vowel [o] consistently elicited N100m peaks 5 ms later than the CORONAL vowel [ø]. In sensor space, an enhancement of RMS peak amplitude for the [ø] vowel by 10 fT (Fig. 2 ) almost attained significance (F 1,20 = 4.12, p < .06). However, the effect was significant in source space that is not influenced by varying head-to-sensor positions: The [ø] dipole source strength, an estimate for the amount of massed neuronal activity, was larger for the [ø] vowel than for the [o] by 25 % or 6 nAm (F 1,16 = 9.36, p < .01). No hemispheric differences in signal power between vowel categories or tasks were apparent. N100m source location and orientation In agreement with previous findings with a more comprehensive set of vowels [ 15 ], the vowel categories [o] and [ø] elicited statistically different centers of activity along the anterior-posterior axis (F 1,16 = 7.73, p < .01), that is, the auditory processing in the DORSAL vowel [o] was reflected by a more posterior ECD location (Fig. 3 ). A difference in source configuration was also evident from a more superior position of the [o] source (F 1,16 = 12.28, p < .01), a more vertical orientation (F 1,16 = 5.81, p < .05) than the [ø] source, and from an angular difference between the two vowel categories in the sagittal plane (i.e. the [o] source was located more posterior and inferior, F 1,16 = 10.91, p < .01) and in the axial plane (i.e. the [o] source was also located more posterior and lateral, F 1,16 = 6.82, p < .05, relative to the [ø] source). None of these effects showed an interaction with hemisphere, but data gained further validity as the right-hemispheric sources were all located more posterior (F 1,16 = 8.88, p < .01), more inferior (F 1,16 = 4.27, p < .06) and were tilted more vertically (F 1,16 = 14.29, p < .01) than their left-hemispheric counterpart. Such a difference is to be expected from previously reported N100 asymmetries between cerebral hemispheres [ 27 - 30 ]. The relative mapping of phonological features of the speech signal [ 14 , 15 ] was not affected by the task-induced shifts of attention. However, shifts of subjects' attentional focus from phonological categorization to identification of the speaker's voice shifted vowel sources as a whole to more posterior and superior locations within the supratemporal plane. Statistically, the speaker categorization task produced more superior (F 1,16 = 4.72, p < .05) and marginally more posterior (F 1,16 = 3.36, p < .10) ECD locations, which was also evident by an angular displacement in the sagittal plane (F 1,16 = 4.6, p < .05). The effect seemed to be driven by changes in the left hemisphere but the task × hemisphere interaction never attained significance (all F < 1). When brain responses were analyzed separately for stimuli spoken by male and female speaker, which yielded satisfying dipole solutions only in 12 subjects, the most striking finding was a consistent speaker × task interaction of the dipole location in both the sagittal plane (F 1,11 = 10.83, p < .01) and the axial plane (F 1,11 = 7.16, p < .03). That is, subjects' attentional focus slightly affected the relative displacement of male and female voice-evoked brain responses: In both the sagittal plane and the axial plane, a significant 4° difference emerged in the phonological categorization task (both p < .05), which vanished in the speaker categorization task. In contrast, as reported above, no such task influence was evident in the relative position of vowel-evoked brain responses. Performance Overall target detection rate was 94.1 %, false alarms occurred in 5.5% of all trials. Responses of the 17 subjects whose brain responses were subjected to magnetic source imaging were analyzed in detail: The phonological categorization task (93.2 ± 3.0 % correct, 4.9 ± 2.2 % false alarms, M ± SEM) and the speaker categorization task (95.0 ± 2.9 % correct, 6.2 ± 3.2 % false alarms) did not differ significantly (one-way repeated measures ANOVAs, all F < 1). Discussion This study was set up to explore potential influences of the attentional focus on the mapping of speech sounds within the auditory cortex. With subject's attention either on the phonological differences or on the speaker difference between vowel stimuli, we mapped the auditory evoked N100m and localized its sources that fitted well with a single dipole per hemisphere. All responses were located in the perisylvian region. Furthermore, the relative distribution of sources indicated an interesting pattern. As hypothesized and expected from previous studies, the fundamental location difference between the sources of the DORSAL vowel [o] source and the CORONAL vowel [ø] [ 15 , 17 ] could be replicated under both attentional conditions. In contrast, the corresponding difference between speaker-dependent sources was subject to task influences. That is, a shift of subjects' attention to a non-phonological acoustic feature, the speaker identity, did not blur the spatial segregation within the speech sound map. In contrast, the [ø] and [o] generators were slightly displaced towards more posterior and more superior locations when subjects focused on speaker identity. In most situations, a listener may automatically extract the phonological invariants from the speech signal in order to access lexical information, for example the meaning of the information inherent in speech. Speaker-dependent features such as pitch and periodicity should not play a crucial role in this phonological decoding process. This is what we mimicked by asking our subjects to detect a certain vowel in a stream of varying speech sounds. However, in cocktail-party-like situations there is the additional demand to attend acoustic properties of certain speech streams or speakers, and we implemented it by asking our subjects to detect a certain voice in a stream of varying speakers. Speaker identification comprises an important but not necessarily orthogonal process to phonological decoding in speech perception: areas in the upper bank of the superior temporal sulcus (STS) have been identified previously [ 31 ] to be voice-selective (as opposed to other environmental sounds), and in many situations the selective tracking of one voice amongst others is a prerequisite for decoding the phonological content of this speaker's utterances. The displacement of dipolar sources seen here may mirror the involvement of additional cortical areas, such as the voice-specialized part in the STS [ 31 ] or pitch-specialized areas in the primary auditory cortex. An additional STS activation would most likely elicit an inferior shift of the dipole sources during speaker categorization. However, a shift into the opposite direction was obtained. This might indicate that the contribution of the voice-specialized part of the STS around 100 ms post-stimulus onset is small compared to other additional cortical areas, such as pitch-specialized areas in the primary auditory cortex. It is now well-established that a finegrained analysis of the speech signal takes place mainly in anterior parts of the supratemporal gyrus [ 17 , 32 - 34 ], thereby anterior of primary auditory areas. Consequently, the activity shift towards more posterior sites we observed in the speaker categorization task strongly argues for an additional involvement of these primary auditory areas. Unfortunately, we cannot dissociate speaker identification processes from pitch processing in the current study. However, pitch differences are among the primary cues dissociating male and female voices, and a clear involvement of auditory core areas in pitch processing has been shown in a recent MEG study focusing on pitch detection mechanisms [ 35 ]. Conclusions Data presented here suggest that the systematic mapping of speech sounds within the auditory cortex is robust under changing attentional demands and not tied to phonological awareness. However, the general shift of activity when a non-phonological speaker categorization must be accomplished shows that speech sound representations are modulated in their locations in a context-dependent manner. Situational demands obviously influence the differential but time-synchronous involvement of specialized neuronal assemblies that contribute to speech sound decoding in a top-down fashion. Hence, the spectrally high-resolving analysis of the incoming speech stream is performed at the same time but in different locations, i.e. in a different mix of cell assemblies than the analysis of speaker-dependent features (such as pitch, periodicity, or other features inherent to voice quality). Further spatially high-resolution brain imaging studies are needed to quantify as to which extent voice-selective areas in the upper bank of the STS [ 31 ] become involved when speaker categorization is accomplished. For the time being, this study increases our understanding of speech sound processing, as it replicates previous findings of an orderly mapping of phonological vowel features and as it shows that changing attentional foci affect the absolute but not the relative distribution of vowel-evoked activity within the auditory cortex. Methods Subjects 22 subjects (11 females, mean age 24.3 ± 4 years, M ± SD) participated in the procedure. All subjects were monolingual native speakers of German. Only right-handers as ascertained by the Edinburgh Handedness Questionnaire [ 36 ] were included. Subjects gave written informed consent and were paid €10 for their participation. Experimental design In an auditory target detection task, subjects listened to randomized sequences of four German natural vowel exemplars: The DORSAL rounded vowel [o] in two exemplars, in one spoken by a male voice and in the other by a female voice, and the CORONAL rounded vowel [ø], also produced by both voices (Fig. 1 ). 200 ms long vowels free of formant transitions were cut out of spoken words, digitized with a 10 kHz sampling rate and faded with 50 ms Gaussian on- and offset ramps. Table 1 summarizes exact pitch and formant frequencies of the four exemplars. Prior to the measurement, individual hearing thresholds were determined for both ears and all four vowel exemplars. Stimuli were presented binaurally with at least 50 dB SL (respective to the vowel exemplar which showed the weakest sensation level, if any differences between exemplars occurred) via a non-magnetic echo-free stimulus delivery system with almost linear frequency characteristic in the critical range of 200–4000 Hz. In a test sequence, subjects repeated vowels aloud and recognized all stimuli correctly, i.e. they distinguished between both vowel categories and voices without difficulty. Binaural loudness was slightly re-adjusted where necessary to ensure perception in the head midline. In the actual measurement, vowel exemplars were presented in two randomized sequences with equal probability and a randomized stimulus onset asynchrony of 1.6 – 2 s. All subjects performed – in a counterbalanced order – two different tasks during these two sequences: In a task A (hereafter called phonological categorization ), subjects had to press a button with their right index finger whenever a given vowel ([o] or [ø], counterbalanced across subjects) occurred, irrespective of the speaking voice. In a task B (hereafter called speaker categorization ), subjects had to press a button whenever a given voice (the male or the female voice, counterbalanced across subjects) uttered a vowel, irrespective of the uttered vowel category. Fig. 1 (lower panel) which clarifies and visualizes the task. That is, in the phonological categorization task, subject's attention was focused on a categorical distinction between speech sounds, [o] or [ø], which closely resembles the tasks applied in most brain imaging studies testing active speech sound processing (e.g. [ 14 , 15 , 37 ]) – a process ubiquitously taking place when decoding running speech. In contrast, the speaker categorization task was intended to shift subject's attention to more general and more basic acoustic properties of the material [ 31 ] presented to accomplish speaker distinction. Data reduction and statistical analyses Data acquisition and analysis, including source modeling, closely followed the procedure described in [ 15 ]: Auditory magnetic fields were recorded using a whole head neuromagnetometer (MAGNES 2500, 4D Neuroimaging, San Diego) in a magnetically shielded room (Vaccumschmelze, Hanau, Germany). Epochs of 800 ms duration (including a 200 ms pre-trigger baseline) were recorded with a bandwidth from 0.1 to 200 Hz and a 687.17 Hz sampling rate. If the peak-to-peak amplitude exceeded 3.5 pT in one of the channels or the co-registered EOG signal was larger than 100 μV, epochs were rejected. Button-presses did not affect the auditory evoked field topography in the N100m time range. We analyzed up to 150 artifact-free vowel responses that remained for both vowel categories [o] and [ø] after off-line noise correction, and averaged them separately for vowel category but across speaker voice. Splitting up vowel conditions into male and female speaker sub-conditions was not possible due to a resulting small number of averages. However, we also performed separate averages and analyses of male and female speaker across vowel categories. In any case, the resulting averages thus contained brain responses to two acoustically variant exemplars which makes results more comparable to our previous studies [ 15 , 17 ]. A 20 Hz lowpass filter (Butterworth 12 dB/oct, zero phase shift) was subsequently applied to the averages. The N100m component was defined as the prominent waveform deflection in the time range between 90 and 160 ms (Fig. 2 ). Isofield contour plots of the magnetic field distribution were visually inspected to ensure that N100m and not P50 m or P200 m were analyzed. N100m peak latency was defined as the sampling point in this latency range by which the first derivative of the Root Mean Square (RMS) amplitude reached its minimum and second derivative was smaller than zero. RMS was calculated across 34 magnetometer channels selected to include the field extrema over the left and the right hemisphere, respectively. Prior to statistical analyses, all brain response latencies were corrected for a constant sound conductance delay of 19 ms in the delivery system. Using the same sets of channels, an equivalent current dipole (ECD) in a spherical volume conductor (fitted to the shape of the regional head surface) was modeled at every sampling point separately for the left and the right hemisphere [ 38 ]. The N100m source parameters were determined as the median of 5 successive ECD solutions in the rising slope of the N100m. The resulting ECD solution represents the center of gravity for the massed and synchronized neuronal activity. To be included in this calculation, single ECD solutions had to meet the following criteria: (i) Goodness of fit greater than .90, (ii) ECD location larger than 1.5 cm in medial-lateral direction from the center of the brain and 3–8 cm in superior direction, measured from the connecting line of the pre-auricular points. Statistical analysis of dependent variables N100m peak latency, amplitude and N100m source generator strength, location and orientation focused on 2 × 2 × 2 repeated measures analysis of variance with repeated factors hemisphere (left vs. right), vowel ([o] vs. [ø]) and task (attend phonology vs. attend speaker). As source location displacements do not appear exactly and exclusively along the Cartesian axes of the source space (cf. [ 21 ]), we additionally calculated differences in the polar angle Φ and the azimuth angle θ which here describe angular displacements in the sagittal and the axial plane, respectively. Authors' contributions J.O., T.E. and C.E. conceived the experiment and drafted the manuscript. J.O. and C.E. prepared the exact experimental setup. J.O. supervised data acquisition, and performed all data and statistical analyses. All authors read and approved the final manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC503386.xml
546194	Patient involvement in medical decision-making and pain among elders: physician or patient-driven?	Background Pain is highly prevalent among older adults, but little is known about how patient involvement in medical decision-making may play a role in limiting its occurrence or severity. The purpose of this study was to evaluate whether physician-driven and patient-driven participation in decision-making were associated with the odds of frequent and severe pain. Methods A cross-sectional population-based survey of 3,135 persons age 65 and older was conducted in the 108-county region comprising West Texas. The survey included self-reports of frequent pain and, among those with frequent pain, the severity of pain. Results Findings from multivariate logistic regression analyses showed that higher patient-driven participation in decision-making was associated with lower odds (OR, 0.82; 95% CI, 0.75–0.89) of frequent pain, but was not significantly associated with severe pain. Physician-driven participation was not significantly associated with frequent or severe pain. Conclusions The findings suggest that patients may need to initiate involvement in medical decision-making to reduce their chances of experiencing frequent pain. Changes to other modifiable health care characteristics, including access to a personal doctor and health insurance coverage, may be more conducive to limiting the risk of severe pain.	Background Persons age 65 years and older commonly endure a multitude of chronic and debilitating conditions which contribute to persistent pain [ 1 ]. Estimates of the prevalence of pain among the community-dwelling elderly range between 25% and 50% [ 2 , 3 ]. Pain has been found to have a substantial effect on health-related quality of life [ 2 ], the use of over the counter and prescription drugs [ 1 , 4 ], and the utilization of medical care [ 5 ]. As the number of elderly persons in the United States rises, more research is needed to determine how the delivery of medical care could be altered to limit the onset of pain and its subsequent burden on health status and the health care system. Increasing patients' involvement in the medical decision-making process is one potentially fruitful means of improving pain management. Several studies suggest that patients, especially those with chronic conditions, who have opportunities to participate in care have more positive health outcomes than those who do not [ 6 , 7 ]. While other studies have pointed out that the positive correlation between patient participation and health outcomes is more suggestive than conclusive, Guadagnoli and Ward have stated that physicians should nevertheless strive to engage their patients in decision-making for humanitarian reasons [ 8 ]. Although patients' participation may improve their health outcomes, the effect can be diminished among elderly patients. Elderly patients, as compared to younger patients, have been shown to be less participatory in medical-decision making [ 9 - 11 ]. Using a longitudinal cohort, the Medical Outcomes Study found that patients older than 75 years were less participatory [ 12 ]. Other studies have also shown that older people tend to exhibit more conversational behaviors [ 13 ], give more socially desirable responses [ 14 ], and defer to physicians' authorities [ 15 ]. The primary objective of the present study was to examine how participation in decision-making was associated with the occurrence of pain among a cohort of community-dwelling elders. In contrast to previous studies, we differentiated two types of participation in decision-making. The first type is physician-driven in which the physician takes the initiative to ask questions and offer choices to patients. The second type is patient-driven, in which the patient takes the initiative to ask questions and express preferences. We hypothesized that stronger physician and patient-driven participation in decision-making would be associated with lower odds of frequent pain and, among those with frequent pain, lower odds of severe pain. We also tested for the effects of other health care factors which might be conducive to pain management, such as tenure with a personal doctor. The findings have implications for how older patients interact with their physicians as well as how physicians and clinic managers organize health services. Methods Sample and setting Data were obtained through a longitudinal, population-based study of community-dwelling elders, the Texas Tech 5000 Survey. The Texas Tech 5000 Survey was conducted in a 108-county region of West Texas, a geographically and ethnically diverse area encompassing approximately half of the state's land mass. The survey has been described in detail elsewhere [ 16 - 21 ]. Briefly, approximately 65,000 households were randomly selected from residential telephone listings and screened to identify a cohort of 5,000 individuals age 65 years and older. Age-qualified individuals were subsequently tested for cognitive impairment using a telephone version of the Mini Mental State Evaluation [ 22 ]. Ninety-three percent of individuals did not have impairment and thus were eligible for participation in the study. Excluding telephone numbers that were never reached, those who refused the cognitive screener, and individuals who failed the cognitive screener, the eligible sample size was 6,942. Two follow-up surveys have since been conducted among the original cohort. Selected questions measuring satisfaction with care and health-related quality of life were included in each wave. To limit respondent burden, most questions were asked only during one wave (the patient and physician-driven participation in decision-making and perceived pain questions were only asked during Wave 3). Of the 6,942 households that were eligible for participation in Wave 1 of the survey, 5,006 persons participated, yielding a baseline response rate of 72%. The data presented here are from 3,135 subjects who participated in all three waves of the survey, yielding an overall response rate of approximately 45%. While some subjects were obviously lost to follow-up, the demographic composition of the study samples remained similar over the study period. The Texas Tech Health Sciences Center Institutional Review Board for the Protection of Human Subjects approved the study. Measures Frequent and severe pain The occurrence and severity of pain were measured in Wave 3 using items developed for and included in two nationally representative surveys of older persons, the Health and Retirement Study (HRS) and Assets and Health Dynamics among the Oldest Old (AHEAD) [ 23 ]. First, the frequency of pain was measured by asking respondents if they were "often troubled with pain?" Responses were categorized to distinguish those persons who were often troubled (referred to hereafter as frequent pain) and those who were not often troubled by pain, as has been done in previous studies [ 2 ]. The severity of pain was assessed among those persons who reported frequent pain through a single item asking, "how bad is the pain most of the time?" Responses were categorized to differentiate those persons with mild or moderate pain versus severe pain. Sociodemographic factors A number of sociodemographic, health care, and health status measures were included. Sociodemographic factors were gender, age (continuous), marital status, educational status (high school graduate vs. less than high school education), and place of residence (urban, rural, and frontier). An urban area is a metropolitan county, or a county with a total population of at least 50,000, whereas a rural area is a county with fewer than 50,000 persons. Rural counties were further classified according to whether they were frontier areas, or counties with fewer than 7 persons per square mile [ 24 ]. Health care factors Health care variables were health insurance coverage, the number of physician visits in the last 6 months, tenure with a personal doctor, an index measuring physician-driven participation in decision-making, and an index measuring patient-driven participation in decision-making. Health insurance coverage was coded as Medicaid, Medicare, Medicare plus other private or government coverage, other private or government coverage, and no insurance. Tenure with a personal doctor was measured using a single question asking if the individual had a personal doctor and, if so, the duration of tenure with the physician (less than 1 year, 1 to 2 years, 3 to 4 years, and 5 or more years). An index of physician-driven participation in decision-making was created using three items taken from the Medical Outcomes Study [ 12 ]. The physician-driven participation in decision-making questions included: 1) How often does your doctor ask you to help make the decision when there is a choice between treatments?, 2) How often does your doctor give you some control over your treatment?, and 3) How often does your doctor ask you to take some of the responsibility? Response options for each item ranged from 0 (never) to 4 (very often). The aggregation of the three items divided by the total number of items produces a score between 0 and 4 with a higher index score indicating greater involvement. For the present data set, the physician-driven participation index had reasonable internal consistency (Cronbach's alpha = 0.69), which was similar to that found in the Medical Care Outcomes Study (Cronbach's alpha = 0.74) [ 12 ]. Three questions adopted from a study of older patients' communication during medical visits [ 25 ] were used to measure patient-driven participation in decision-making. These questions included: 1) How often do you write out a list of symptoms, complaints, and medications before visiting a doctor?, 2) How often do you express preferences for tests, medications, and treatments?, and 3) How often do you call to clarify information or report symptoms or side effects after a visit? As was the case for the physician-driven index, the patient-driven participation index ranges between 0 and 4 with a higher score indicating greater involvement. The Cronbach's alpha for the patient-driven participation index was 0.58. Health status Overall health status (categorized as excellent, very good, good, and fair or poor) was measured using a general health item from a brief health-related quality of life instrument (the SF-12) [ 26 ]. Mental health status was assessed with the mental component score (MCS) of the SF-12. Additional health status variables included whether the individual had ever been diagnosed with arthritis and the number of additional comorbid conditions (categorized as none, one, two, and three or more). Statistical analysis Chi-square tests were first conducted to determine whether there was an association between each categorical sociodemographic, health care, and health status factor and frequent and severe pain. T-tests were conducted to determine if there was a difference in each continuous sociodemographic, health care, and health status factor between individuals with and without frequent pain and individuals with and without severe pain. Next, multivariate logistic regression analyses were conducted to determine if physician or patient-driven participation in decision-making were associated with the odds of frequent pain and, among those with frequent pain, the odds of severe pain. The potential for multicollinearity between the covariates was assessed by calculating their variance inflation factors; no problems with multicollinearity were found. Results Description statistics for individuals with frequent pain A total of 1,333 (42.5%) of the 3,135 survey participants had frequent pain. Table 1 presents percentages for frequent pain by categorical sociodemographic, health care, and health status variables. Several categorical sociodemographic variables were significantly associated with frequent pain. Frequent pain was less common among males (compared to females) and among married persons (compared to single persons). Frequent pain was more common among persons with less than a high school degree as compared to those with at least a high school degree. Health insurance was insignificant, but tenure with a personal doctor was associated with frequent pain. Specifically, pain was least common among individuals with no personal doctor, compared to those who had a personal doctor. Frequent pain was more common among persons with more comorbid diseases or conditions (compared to those with none), those with arthritis (compared to those without arthritis), and those with poorer self-rated general health. As shown in Table 2 , persons with frequent pain had a higher number of physician visits in the previous six months but lower physician and patient-driven participation than those without frequent pain. Moreover, those with frequent pain had worse (lower) mental component scores (MCS) than those without frequent pain. Table 1 Prevalence of frequent and severe pain by categorical independent variables Sociodemographics N = 3,135 Frequent Pain % Severe Pain % Gender Male 952 35.7 1 6.3 3 Female 2,183 45.5 10.4 Race/Ethnicity White 2,678 42.5 8.4 1 Hispanic 327 42.5 11.6 Other 130 43.9 19.2 Education < high sch. grad. 785 45.9 3 13.3 1 High school grad. 2,350 41.4 7.8 Marital Status Married 1,661 40.8 3 8.1 Not married 1,474 44.5 10.3 Rural / urban residency Urban 1,720 43.9 9.6 Rural, non-frontier 1,072 40.8 8.7 Frontier 343 41.1 8.5 Income <$10,000 491 47.9 1 14.3 2 $10–20,000 658 47.7 10.3 $21–30,000 505 43.0 7.7 $31,000 and higher 892 36.0 5.7 Health care Insurance Uninsured 93 46.2 21.5 1 Medicare 917 41.8 9.2 Medicaid 341 43.7 12.6 Medicare+other ins. 1,498 43.4 7.9 Other private/gov. ins 286 37.8 7.7 Tenure with doctor No personal doctor 426 37.8 1 10.6 Less than 1 yr 227 44.9 9.2 1–2 yrs 414 45.7 10.1 3–4 yrs 454 42.3 8.4 5 or more yrs 1,614 42.7 8.7 Health status No. of comorbidities 0 1,690 37.2 3 6.3 2 1 874 47.1 10.5 2 384 48.7 14.1 3 or more 187 56.7 18.7 Ever diagnosed with arthritis Yes 1,988 55.0 1 12.8 2 No 1,147 20.8 2.9 General health status Excellent 316 15.2 3 2.2 3 Very good 720 31.1 5.6 Good 1,040 41.0 3.1 Fair 687 55.8 13.8 Poor 360 69.2 29.2 1 p < 0.0001, 2 p < 0.01, 3 p < 0.05 note: Analyses of severe pain were limited to those with frequent pain. Table 2 Means and standard deviations for continuous independent variablesby frequent and severe pain Frequent pain Severe pain Yes No Yes No Sociodemographics Mean age (SD) 75.4(6.3) 75.3(6.3) 75.7(6.7) 75.3(6.3) Health care Mean no. physician visits (SD) 6.3(8.5) 1 3.9(6.1) 8.0(9.5) 1 5.8(8.1) Mean physician-driven participation index (SD) 1.8(1.2) 3 1.9(1.2) 1.9(1.2) 3 1.7(1.2) Mean patient-driven participation index (SD) 2.4(1.0) 1 2.7(1.0) 2.4(1.0) 2.4(1.1) Health status Mean SF-12 mental component score (SD) 52.5(9.9) 1 54.8(7.4) 48.7(12.1) 1 53.6(9.0) 1 p < 0.0001, 2 p < 0.01, 3 p < 0.05 note: Analyses of severe pain were limited to those with frequent pain. Description statistics for individuals with severe pain Among the 1,333 individuals with frequent pain, a total of 287 (21.5%) had severe pain. As shown in Table 1 , severe pain was less common among males (compared to females) but more common among other races/ethnicities (compared to non-Hispanic whites), persons with less than a high school degree (compared to at least a high school degree), and persons with lower household income. Severe pain was most common among individuals without health insurance. It was more common among persons with more comorbid diseases or conditions (compared to those with none), those with arthritis (compared to those without arthritis), and those with poorer self rated general health status. As shown in Table 2 , those with severe pain had more physician visits and higher physician participation in decision-making. Those with severe pain had lower or worse mental component scores and a higher number of physician visits in the previous six months (compared to those without severe pain). Multivariate analyses of the odds of frequent pain Findings from multivariate logistics analyses are shown in Table 3 . Males had lower odds (OR, 0.81; 95% CI 0.67, 0.98) of frequent pain than females. Race/ethnicity was not significantly associated with frequent pain. Compared to urban residents, those residing in a rural area had lower odds (OR, 0.77; 95% CI 0.64, 0.92) of frequent pain than urban residents. Income, marital status, and frontier residence were not significantly associated with the odds of frequent pain. Individuals who had more physician visits in the previous six months had a higher odds of frequent pain (OR, 1.02; 95% CI 1.01, 1.04). Table 3 Multivariate logistic regression of sociodemographic, health care, and health status factors on frequent and severe pain Variable (reference group) Frequent pain OR (95% CI) Severe pain OR (95% CI) Sociodemographics Age 0.99 (0.97, 1.00) 1.00 (0.98, 1.02) Male (vs. female) 0.81 (0.67, 0.98) 3 0.78 (0.54, 1.14) Race/ethnicity Hispanic (vs. non-Hispanic white) 0.74 (0.54, 1.01) 0.74 (0.44, 1.26) Other (vs. non-Hispanic white) 0.84 (0.56, 1.26) 2.28 (1.24, 4.21) 2 Less than high school grad. (vs. grad.) 0.97 (0.78, 1.20) 1.08 (0.76, 1.55) Married (vs. single) 0.92 (0.77, 1.10) 1.00 (0.72, 1.39) Residence Rural (vs. urban) 0.77 (0.64, 0.92) 3 0.89 (0.65, 1.23) Frontier (vs. urban) 0.86 (0.66, 1.12) 0.78 (0.48, 1.28) Income $10–20,000 (vs. < $10,000) 1.06 (0.81, 1.39) 0.81 (0.52, 1.26) $21–30,000 (vs. < $10,000) 1.02 (0.75, 1.38) 0.96 (0.57, 1.62) $31,000 and higher (vs. < $10,000) 0.90 (0.67, 1.20) 0.87 (0.52, 1.46) Missing (vs. < $10,000) 0.95 (0.72, 1.25) 0.99 (0.63, 1.56) Health care Insurance Medicare (vs. uninsured) 0.69 (0.41, 1.16) 0.41 (0.19, 0.86) 3 Medicaid (vs. uninsured) 0.72 (0.44, 1.18) 0.35 (0.17, 0.71) 2 Medicare plus other. (vs. uninsured) 0.85 (0.52, 1.39) 0.31 (0.15, 0.64) 2 Other private/gov. ins. (vs. uninsured) 0.64 (0.37, 1.09) 0.34 (0.15, 0.76) 2 No. of physician visits past 6 months 1.02 (1.01, 1.04) 2 1.02 (1.00, 1.03) Physician-driven participation index 0.99 (0.91, 1.06) 1.14 (0.99, 1.30) Patient-driven participation index 0.82 (0.75, 0.89) 3 0.93 (0.80, 1.09) Tenure with doctor Less than 1 year (vs. no personal doctor) 0.92 (0.64, 1.33) 0.51 (0.27, 0.98) 3 1–2 years (vs. no personal doctor) 0.95 (0.70, 1.30) 0.74 (0.43, 1.25) 3–4 years (vs. no personal doctor) 0.87 (0.64, 1.18) 0.56 (0.33, 0.96) 3 5 or more years (vs. no personal doctor) 0.92 (0.71, 1.19) 0.65 (0.42, 0.99) 3 Health status No. of comorbidities 1 (vs. none) 1.06 (0.87, 1.27) 1.18 (0.84, 1.66) 2 (vs. none) 0.96 (0.74, 1.25) 1.50 (0.99, 2.28) 3 or more (vs. none) 1.01 (0.71, 1.43) 1.51 (0.90, 2.50) Ever diagnosed with arthritis (vs. never) 3.62 (3.03, 4.33) 1 1.54 (1.01, 2.35) 3 SF-12 mental component score 0.99 (0.98, 1.00) 0.98 (0.97, 1.00) General health status Excellent (vs. poor) 0.14 (0.09, 0.22) 1 0.49 (0.20, 1.19) Very Good (vs. poor) 0.30 (0.21, 0.41) 1 0.24 (0.14, 0.42) 1 Good (vs. poor) 0.41 (0.31, 0.54) 1 0.33 (0.21, 0.50) 1 Fair (vs. poor) 0.69 (0.52, 0.92) 2 0.56 (0.39, 0.81) 3 1 p < 0.0001, 2 p < 0.01, 3 p < 0.05 note: Analyses of severe pain were limited to those with frequent pain. Physician-driven participation was not significantly associated with the odds of frequent pain. However, elders with higher patient-driven participation had lower odds (OR, 0.82; 95% CI 0.75, 0.89) of frequent pain, confirming our hypothesis that persons who take a more active role in their medical treatment are less likely to experience pain. Individuals who had been diagnosed with arthritis at some point in their lives had a higher odds of frequent pain (OR, 3.62; 95% CI 3.03, 4.33) than those without arthritis. Those who rated their general health as excellent (OR, 0.14; 95% CI 0.09, 0.22), very good (OR, 0.30; 95% CI 0.21, 0.41), good (OR, 0.41; 95% CI 0.31, 0.54), and fair (OR, 0.69, CI 0.52, 0.92) had lower odds of frequent pain than those who rated their health as poor. Multivariate analyses of the odds of severe pain Among those with frequent pain, there were no gender difference in the odds of severe pain. Persons of other race/ethnicity (primarily Black/African Americans) had a higher odds (OR, 2.28; 95% CI 1.24, 4.21) of severe pain than non-Hispanic whites. Income, marital status, rural residence, and frontier residence were not significantly associated with the odds of severe pain. The number of physician visits was not significantly associated with severe pain. However, having insurance had a significant impact on the odds of severe pain. Compared to those without health insurance coverage, those with Medicare (OR, 0.41; 95% CI 0.19, 0.86), Medicaid (OR 0.35; 95% CI 0.17, 0.71), Medicare plus other private or government coverage (OR 0.31; 95% CI 0.15, 0.64), and those with other private or government coverage (OR 0.34; 95% CI 0.15, 0.76) had a significantly lower odds of severe pain. Although physician and patient-driven participation were not significantly related to the odds of severe pain, tenure with one's personal doctor was a significant factor. Elders who had been seeing their doctor for less than 1 year (OR, 0.51; 95% CI 0.27, 0.98), 3–4 years (OR, 0.56; 95% CI 0.33, 0.96), or 5 or more years (OR, 0.65; 95% CI 0.42, 0.99) had lower odds of severe pain than elders who had no personal doctor. As expected, several health status measures were also significant. Those who had arthritis had higher odds of severe pain (OR, 1.54; 95% CI 1.01, 2.35) than those without arthritis. Finally, individuals who rated their general health as very good (OR, 0.24; 95% CI 0.14, 0.42), good (OR, 0.33; 95% CI 0.21, 0.50), and fair (OR, 0.56; 95% CI 0.39, 0.81) had lower odds of severe pain than those who rated their health as poor. Discussion We found that patient-driven participation in decision-making was associated with lower odds of frequent pain, which is supported by previous research indicating that adult patients who are more actively engaged in their treatment have greater reductions in symptoms and improvement in health status [ 27 ], better psychological outcomes [ 28 ], and higher satisfaction with health care [ 29 ]. Thus, to delay or prevent the development of frequent pain, elderly patients may need to initiate discussions about symptoms with their physicians when they first experience them. However, because many elderly patients often defer to the doctor to initiate involvement in medical decisions [ 9 - 15 ], this may prove to be a difficult task. Little research has investigated how to promote active participation in medical care decision-making, but one prior study which involved sharing of a patient's medical record and the delivery of brief education about his/her disease prior to a physician visit demonstrated that patient involvement in decision making increased [ 30 ]. While neither physician nor patient-driven participation in decision making were significantly associated with pain severity, another factor related to the strength of the doctor-patient relationship was significant. In the present study, having a personal doctor, no matter how long the tenure of the relationship was, reduced the odds of severe pain. The finding of a beneficial effect of having a personal doctor, at least in terms of the severity of pain, is consistent with prior studies which have shown that having a usual source of care is positively correlated with an individual's access to the health care system [ 31 - 33 ], satisfaction with medical care [ 34 ], and promotion of proper medication use [ 35 ]. A usual, personal doctor undoubtedly has a more thorough knowledge of a patient's medical history and problems, which could enable him/her to more effectively manage pain treatment and coordinate care with specialists, if necessary. If this is the case, managers and leaders of physician clinics that have a high mix of elderly patients should ensure that patients can visit a regular doctor to promote better pain management. In addition to having a personal doctor, access to any type of health insurance coverage was associated with the odds of severe pain. Approximately 12 percent of persons in the study sample reported that they had no health insurance coverage at all, including Medicare, and 21.5% of those without insurance had severe pain. The percentage of patients in this group with severe pain was nearly 2 times higher than the percentage of patients in the other insurance categories. Many older persons may not be eligible for public health insurance because they have not contributed to the social security system for a minimal amount of time. This may be particularly common in the southwestern United States where there are larger numbers of Hispanic immigrants. Expansion of health insurance coverage to this group could improve their ability to visit physicians and other health providers when they experience pain and, ultimately, lead to better pain management. Further research is warranted to more clearly elucidate how characteristics of different health insurance plans, such as gatekeeping and cost sharing, affect access to physician services for pain treatment. Several demographic indicators were also significantly associated with frequent and severe pain. The gender differences beckon the question of whether medical care providers treat older women's pain less effectively or appropriately than men's. No differences were found between Hispanics and non-Hispanic whites, but other races (the majority of whom were Black/African American) had higher odds of severe pain than non-Hispanic whites. However, because of the heterogeneity of the other racial category, it is difficult to discern which particular racial groups experience severe pain. Research which includes greater numbers of other racial categories is thus warranted. In regard to health status, the results support that individuals with three or more comorbid diseases have a relatively higher odds of frequent pain and severe pain than those with no comorbid diseases. One disease, arthritis, was of particular interest and therefore was treated as a separate variable. Almost two-thirds of the subjects had arthritis, which is not unexpected for persons age 65 and older. Persons with arthritis had a much higher odds (over 3 times the odds) odds of frequent pain than individuals without arthritis. Moreover, those with arthritis had approximately 1.5 times the odds of severe pain. The magnitudes of these associations imply that efforts to more effectively treat arthritis could lead to improvements in pain management. While the present study has contributed to our understanding of the relationship between doctor-patient interactions and persistent pain, it is not without several limitations. Because the study was cross-sectional in design, it is impossible to infer any causal relationships. Although the pain measures were adapted from a nationally representative cohort study of older persons [ 23 ], they may not adequately reflect the frequency, duration, and severity of pain. The generalizability of the findings may be limited to regions of the southwestern United States that are similar in geographic and ethnic makeup, such as Texas, New Mexico, Colorado, Arizona, and California. However, we suspect that the associations found in the present study would hold true among elders residing in other parts of the United States. In summary, future studies should employ longitudinal designs, include more detailed measures of pain and be conducted among other populations. Conclusions Despite these potential limitations, the present study suggests that several strategies could be implemented to limit the incidence and severity of pain among community-dwelling elders. Health policy makers and insurance companies might implement new reimbursement schemes to encourage visits to a personal physicians in order to improve pain and other health outcomes. Managers of physician clinics should consider organizing practices to ensure that older patients are able to make timely appointments with a personal provider. Finally, patients themselves could help reduce their chances of having frequent pain by becoming more involved in their care. These are just a few examples of how changes to the organization and delivery of care might affect pain-related health outcomes. Future research should evaluate how a range of physician characteristics (e.g. specialty and age), physician clinic characteristics (e.g. solo or group practice), insurance characteristics (e.g. HMO, PPO, or FFS coverage), and patient characteristics (e.g. trust in physician) influence pain and pain treatment. Competing interests The author(s) declare that they have no competing interests. Authors' contributions TFB conceived the overall study, performed the statistical analyses, and led the drafting of the manuscript. KTX assisted with the study design, interpretation of statistical findings, and drafting of the manuscript. JH assisted with interpretation of the findings and drafting of the clinical implications. GK contributed to data management, statistical analyses, and drafting of the methods section. All authors read and approved the manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC546194.xml
515309	The planetary biology of cytochrome P450 aromatases	Background Joining a model for the molecular evolution of a protein family to the paleontological and geological records (geobiology), and then to the chemical structures of substrates, products, and protein folds, is emerging as a broad strategy for generating hypotheses concerning function in a post-genomic world. This strategy expands systems biology to a planetary context, necessary for a notion of fitness to underlie (as it must) any discussion of function within a biomolecular system. Results Here, we report an example of such an expansion, where tools from planetary biology were used to analyze three genes from the pig Sus scrofa that encode cytochrome P450 aromatases–enzymes that convert androgens into estrogens. The evolutionary history of the vertebrate aromatase gene family was reconstructed. Transition redundant exchange silent substitution metrics were used to interpolate dates for the divergence of family members, the paleontological record was consulted to identify changes in physiology that correlated in time with the change in molecular behavior, and new aromatase sequences from peccary were obtained. Metrics that detect changing function in proteins were then applied, including K A /K S values and those that exploit structural biology. These identified specific amino acid replacements that were associated with changing substrate and product specificity during the time of presumed adaptive change. The combined analysis suggests that aromatase paralogs arose in pigs as a result of selection for Suoidea with larger litters than their ancestors, and permitted the Suoidea to survive the global climatic trauma that began in the Eocene. Conclusions This combination of bioinformatics analysis, molecular evolution, paleontology, cladistics, global climatology, structural biology, and organic chemistry serves as a paradigm in planetary biology. As the geological, paleontological, and genomic records improve, this approach should become widely useful to make systems biology statements about high-level function for biomolecular systems.	Background The emergence of complete genomes for many organisms, including humans, has created the need for hypotheses concerning the "function" of specific genes that encode specific proteins. While "function" is interpreted by different workers in different ways [ 1 ], Darwinian theory (by axiom) requires that the term be connected to fitness; natural selection is the only mechanism admitted by theory to generate functional behavior in a living system, macro or molecular. This, in turn, implies that the hypotheses about function have a "systems" component, including the interaction of the protein with other proteins, their impact on the physiology (defined broadly) of the cell and organism, and the consequences of physiology in a changing ecosystem in a planetary context [ 2 ]. Systems hypotheses can be supported by information from many areas. Geology, paleontology, and genomics, for example, provide three records that capture the natural history of past life on Earth. At the same time, structural biology, genetics, and organic chemistry describe the structures, behaviors and reactivities of proteins that allow them to support present life. It has been appreciated that a combination of these six types of analysis provides insights into functional behavior of proteins that cannot be provided by any of these alone [ 2 ]. Over the long term, we expect that the histories of the geosphere, the biosphere, and the genosphere will converge to give a coherent picture showing the relationship between life and the planet that supports it. This picture will be based, however, on individual cases that serve as paradigms for making the connection. The aromatase family of proteins offers an interesting system to illustrate the power of this combination as a way to create hypotheses regarding protein function within a system [ 3 ]. These hypotheses are not "proof", of course, but are limiting in genomics-inspired biological experimentation, now that genomic data themselves are so abundant. Aromatases are cytochrome P450-dependent enzymes that use dioxygen to catalyze a multistep transformation of an androgenic steroid (such as testosterone) to an estrogenic steroid (such as estradiol) (Figure 1 ). The protein plays a key role in normal vertebrate reproductive biology–a role that appears to have arisen before fish and tetrapods (land vertebrates, including mammals) diverged some 375 million years ago [ 4 ]. Aromatase is important in modern medicine as well, especially in breast and other hormone-dependent cancers [ 5 ]. Different numbers of aromatase genes are found in different vertebrates. Two aromatase genes are known in teleost fish [ 6 , 7 ]. Only a single gene is known in the horse [ 8 ], rat [ 9 ], and mouse [ 10 ]. Cattle have both a functional gene and a pseudogene built from homologs of exons 2, 3, 5, 8, and 9 of their functional gene; these are interspersed with a bovine repeat element [ 11 , 12 ]. In several mammalian species, including humans and rabbits, a single gene yields multiple forms of the mRNA for aromatase in different tissues via alternative splicing [ 13 - 16 ]. A still different phenomenology is observed in the pig ( Sus scrofa ). Three different mRNA molecules had been reported in different tissues from pig [ 17 - 21 ]. Compelling evidence then emerged that the three variants of mRNA identified in cDNA studies arose from three paralogous genes [ 22 ], rather than from a single gene differentially spliced [ 23 ]. This implies that the three aromatase paralogs in pigs arose via gene duplications relatively recent in geologic time. Hypotheses relating to the function of the three aromatase paralogs depend in part on when those duplications took place. If they were very recent, the three genes might have helped pigs adapt to domestication. If they pre-dated the divergence of pigs and fish [ 6 ], they may have different roles that are very fundamental to reproductive endocrinology in vertebrates. We apply here a series of tools to generate better hypotheses concerning the aromatase family of paralogs in swine. Results One strategy useful for understanding the function of genes correlates events in their molecular evolution with events occurring in the history of other genes in the same and/or neighboring lineages, and with events recorded in the geological and paleontological records [ 2 ]. We incorporated a tool to date the divergence of two or more genes through an analysis of transitions at synonymous sites of two-fold redundant coding systems, where the encoded amino acid has been conserved [ 24 ]. This analysis exploits the approach-to-equilibrium kinetic behavior displayed by these sites. The analysis yields a transition redundant exchange (TREx) distance for any gene pair where the synonymous sites have not equilibrated. To calibrate the silent TREx clock, inter-taxa histograms relating pig ( Sus scrofa ) and ox ( Bos taurus ) were constructed for transitions at the silent sites of two-fold redundant codon systems where the encoded amino acid was conserved between the species [ 24 ]. The major peaks associated with the separation of these two lineages was observed at f 2 = 0.87, corresponding to a TREx distance of kt = 0.332. As the fossil record constrains the date of divergence of these two lineages to be 60 ± 5 Ma [ 25 - 27 ], and the codon biases in modern Sus scrofa and Bos taurus project an equilibrium value for f 2 = 0.54 [ 24 ], the rate constants for transitions at the TREx silent sites were estimated to be ca. 2.8 × 10 -9 transitions/silent site/year during the time interval that separates these lineages. Analogous f 2 values were then obtained for other vertebrate aromatase pairs, including fish vs. tetrapods ( f 2 = 0.56), birds versus mammals ( f 2 = 0.612), primates versus ungulates ( f 2 = 0.823), and horses versus artiodactyls ( f 2 = 0.828). Assuming a time-invariant single lineage first order rate constant of 3.6 × 10 -9 changes/site/year and an equilibrium f 2 of 0.54, the corresponding dates of divergence are calculated to be 435, 258, 67, and 65 Ma respectively, with the oldest dates being the least precise. The last three of these dates of divergence are similar to those suggested by the paleontological record [ 28 ], within the error of the calculation, which reflects the modest number of characters used to calculate the f 2 values. A tree for the artiodactyl lineage was constructed from the corresponding TREx distances (Figure 2 ). This was found to be consistent with the tree constructed from other metrics. The TREx clock is not widely used. It may, however, provide more accurate dates in regions where synonymous transitions have not equilibrated than conventional clocks that combine data from synonymous transitions and synonymous transversions, or from non-synonymous changes. A comparison of different clocks will be provided in detail elsewhere (Benner et al. , in preparation). Briefly, the rate constants for transitions and transversions are more different than the two rate constants for purine-purine and pyrimidine-pyrimidine transitions. Further, nucleotide frequencies can be used to calibrate the end equilibrium points for two-fold redundant codon systems directly, and this permits an "approach to equilibrium" formalism, well known in chemical kinetics, to be applied [ 24 , 29 - 31 ]. From the tree, the TREx distances from the ancestor of fetal and placental aromatase to the modern enzymes are 0.113-0.079 (using an endpoint of 0.54 to reflect equilibration at the silent sites), corresponding to a range in the time of divergence of 26–38 Ma. The TREx distances from the divergence of all of the porcine aromatases and the modern forms ranges from 0.082–0.116, corresponding to dates of divergence in the range of 27–39 Ma. This suggests that the three aromatase paralogs diverged in the late Eocene to mid Oligocene. To further correlate the duplication of the genes with the fossil record, genomic DNA was analyzed from relatives of Sus scrofa . Both peccary and babirusa seminal plasma ( Tayassu pecari , from the Center for Reproduction of Endangered Species, Zoological Society of San Diego; Babyrousa babyrussa , from the Bronx Zoo, New York) was probed by PCR (Polymerase Chain Reaction) amplification using exon 4-specific primers [ 32 ]. Bands having the sizes expected for the corresponding aromatases were observed by agarose gel electrophoresis. Based on sequence similarity, two isoforms of aromatase were obtained from both peccary and babirusa as clones derived from the PCR products (Figure 3 ). This establishes that at least one of the duplications occurred before the Tayassuidae (the peccaries) diverged from the Suidae (the true pigs) ca. 35 Ma [ 33 , 34 ]. These data are consistent with an evolutionary model that holds that the ancestor of pig and oxen (approximated in the fossil record by Diacodexis , from the early Eocene ca. 55 Ma) [ 35 ] contained a single aromatase gene, and that the paralogous genes in pig arose some 20 million years later. This suggests that the paralogs in pig can be explained neither in terms of the fundamentals of vertebrate reproductive endocrinology, nor as a consequence of swine domestication. This does, however, suggest that the emergence of the aromatase paralogs was approximately contemporaneous with the emergence of a litter in the Suoidea larger than that found in the ancestral artiodactyl condition. While ruminant and camelid artiodactyls have only one-two young per litter, suoids in general have at least two young per litter (as seen in peccaries) and most suines (true pigs) routinely have three-four young (up to 12 in the domestic pig, Sus ). Note that there has long been the tacit assumption that large litters in suoids represent the primitive artiodactyl condition. Large litters are primitive for mammals in general, and because suoids are plesiomorphic in some anatomical conditions relative to other artiodactyls (e.g., short legs, retention of four digits, bunodont cheek teeth), they have been assumed to be plesiomorphic in other respects. Other data suggest that small litters are in fact the primitive artiodactyl condition. Tragulids (mouse deer or chevrotains) are surviving small, primitive ruminants that are not too dissimilar from Diacodexis in body form, but only have one-two young per litter. Additionally, fossil record data on pregnant oreodonts (an extinct group probably related to the ruminant/camelid artiodactyl lineage, but with a suoid-like plesiomorphic postcranial morphology) shows that they also only had one-two young [ 36 , 37 ]. A cladogram of the Artiodactyla (Figure 4 ) illustrates the probable acquisition of multiparous versus uniparous reproductive strategies, and places the character of litters with typically more than two members emerging just before the divergence of Tayassuidae and Suidae. The approximate correlation in time of the aromatase divergence in Suoidea with the enlargement of litters in Suoidea suggests, as a hypothesis, that the two are functionally related. To expand on this hypothesis, we sought genomic signatures of functional change within the aromatase paralogs. The number of non-synonymous changes in the gene divided by the number of the synonymous changes, normalized for the number of non-synonymous and synonymous sites (the K A /K S value), strongly suggests functional change when the value is significantly greater than unity [ 38 , 39 ], and is also an indicator of hypothetical functional change when the value is high on a branch of a tree relative to other branches of the same tree [ 40 - 43 ]. K A /K S values were reconstructed for individual branches of the evolutionary tree derived from the Darwin bioinformatics workbench (see Methods) using a distance matrix and ancestral states constructed by the method of Messier and Stewart [ 39 ]. The typical branch in the aromatase evolutionary tree has a K A /K S value of 0.35. A higher K A /K S value of 0.85 is found in the episodes of evolution near when the pig aromatases diverged. While a K A /K S value of 0.85 does not require the conclusion that positive selection occurred during the emergence of these aromatase paralogs, an inference based on the magnitude of K A /K S in one branch, relative to the K A /K S value for typical branches [ 40 - 43 ], suggests that adaptive changes occurred during the duplications of the aromatase genes in pigs. A complete maximum likelihood analysis of the aromatase gene family was performed using the PAUP and PAML programs. The resulting tree, generated in PAUP, is shown in Figure 5 , with parameters estimated using PAML. Once more, the generation of paralogs in the pig was found to have occurred after the divergence of pigs from oxen. A high K A /K S value (0.93) was again found in the divergence of the swine isoforms on the branch leading to the ancestor of the placental and embryonic enzymes following their divergence from the pig ovarian enzyme. The distribution of substitutions along this branch is consistent with altered functional constraints for the placental and embryonic enzymes compared with their extinct and extant counterparts (Tables 1 and 2 ) [ 44 ]. We correlated the episode of rapid sequence change during the emergence of the embryonic and placental paralogs with the structural biology of aromatase. A homology model of aromatase was built from progesterone 21-hydroxylase from rabbit liver (coordinates from PDB file 1DT6) [ 45 ], a homologous cytochrome P450-dependent monooxygenase. Residues undergoing replacement during the episodes represented by branches in Figure 5 (branches 1–3) are highlighted in color on the 3D model using a program in prototype with HyperChem (Figure 6 ). Multiple features within the pattern of amino acid replacement were apparent. First, the sites accepting amino acid replacements in the branches with low K A /K S values (as represented by branch 2 in Figure 5 ) were typically scattered without any obvious pattern over the surface of the protein. This is expected for neutral drift, although an adaptive role for these replacements is not excluded by this analysis. In contrast, the distribution of sites accepting amino acid replacements during the episode of rapid sequence evolution of branch 1 (as indicated by a relatively high K A /K S value) involving pig paralogs was not random over the protein surface. Rather, the sites are clustered near the substrate binding pocket, and in a region of the surface believed to contact the co-reductant protein, as identified by mutagenesis experiments in the homolog [ 46 , 47 ]. The clustering of amino acid replacements near a substrate binding site during an episode of rapid sequence evolution suggests that the substrate specificity of the protein might be changing in correlation with a change in the detailed physiological role of the protein. Recent reports suggest that the substrate and product specificities of the placental and embryonic enzymes are indeed different from those of the ovarian enzyme [ 23 , 48 - 50 ]. Further, synthesis of estrogen by the ovarian enzyme is more dependent on the structure of the co-reductant than is the placental enzyme [ 51 ]. Our in silico analyses rationalize these experimental observations from a structural perspective. The coupling of an evolutionary analysis to a crystallographic analysis suggests that the amino acid changes are functionally significant. Discussion Today, natural history holds some of the most intellectually challenging conundrums to ever fascinate the human mind. Further, natural history offers biological chemists the opportunity to place broad biological meaning on the detailed analysis of the structure reactivity of isolated biological molecules studied in a reductionist setting. To do so, however, natural history must be connected to the physical and molecular sciences, both in subject matter and in culture. In part to make this connection, natural historians have sought to change the research paradigm in their field to favor quantitative data directed towards the "proof" of hypotheses over "story telling". Proving hypotheses is difficult in natural history ( pace the philosophical reality that no significant statement in empirical science can ever be said to be "proven"). The events of interest (such as the extinction of dinosaurs) are frequently distant in time, or require a passing of time (as for speciation), making them difficult to reproduce in a laboratory. The scale of the concepts involved (species, environments, planets) also does not lend these concepts to laboratory models and laboratory-controlled tests. Further, a reductionist approach, even when available, will not necessarily generate data that are relevant to the big issue that concerns the natural historian. The emphasis on data and proof has ameliorated the worst excesses of storytelling in natural history, with enormous positive impact. Just as natural historians were purifying their field of storytelling, however, whole genome sequences began to emerge. By dramatically increasing the quantity of chemical data concerning the molecular structures of proteins, genomics changed the limiting steps in biochemical and biomedical research. No longer was the typical researcher attempting to solve an organic chemical or biotechnological question (What is the sequence of my protein? How do I express it at high levels to get the sequence?) for a protein that had been selected for functional reasons. Today, the typical researcher knows the structure of many proteins, and wishes to select one for expression and study based on a hypothesis about its potential function. Here, the fact that any definition of function, which must make reference to fitness, requires some systems, ecological, or planetary context, makes the natural historian a natural source of hypotheses. Their full reductionist armamentarium is available in the laboratory to test and explore any hypothesis that the natural historian might provide. The biomedical researchers may like some guidance from the natural historian to narrow the broad selection, or to shorten the random walk, if only slightly. For this purpose, the forswearing by natural historians of storytelling has come at a most inopportune time. To the modern natural historian, creating hypothesis can easily be regarded as "storytelling". They are reluctant to do so, and may criticize as atavistic colleagues who do. This has created a vacuum in the scientific community. Very few laboratories exist that can draw upon an expertise in natural history to generate stories that create hypotheses for the researcher working in experimental biochemistry and molecular biology. This article is designed in part to illustrate how this vacuum might be filled. Here, we do not just tell a story based on natural history, or even a story based on natural history supplemented with physiology and molecular sequence data. Rather, we show how the addition of other data, including data from X-ray crystallography, can make a story sufficiently rich that it can be viewed as being internally consistent with a wide range of independent data drawn from independent sources. This creates a hypothesis that is more than a story, even if it is less than proven. With aromatase, the congruence of our different analyses makes a compelling suggestion that the three aromatase paralogs in pigs arose by two duplication events in the late Eocene or early Oligocene. The emergence of the aromatase paralogs corresponded approximately in time to the emergence of larger litter size in suines. This implies that the two duplication events are functionally related to the larger litter sizes. This inference is consistent with the physiological impact of estrogen synthesis by these paralogs in Sus . Steroid production by the porcine embryo is tightly controlled by the transient expression of aromatase and 17-hydroxylase (P450C17) between days 10 and 13 [ 20 , 21 , 52 ]. In contrast, estrogen synthesis by the equine embryo begins as early as day 6 and increases with embryo age and diameter [ 52 ]. The estrogen produced by the pig embryonic aromatase is believed to have an impact on the mobility, spacing, and implantation of the concepti [ 52 - 56 ]. Adequate spacing would appear to be required to manage a larger litter. This is consistent with a structural biological analysis that correlates specific amino acid replacements with specific changes in the substrate and product specificity of the protein [ 57 ]. Interestingly, the substrate specificity of human aromatase is reported to be more similar to that displayed by the pig placental enzyme than the ovarian form [ 48 , 49 ]. This is an unexpected similarity given that our evolutionary analysis suggests a change in biochemical function along the fetal/placental branch in the Suidae. It should be noted that the hypothesis is supported by the combination of data that individually would not have strength past storytelling. Thus, the K A /K S ratio of 0.93 would not, by itself, compel any particular interpretation. Its implications are greater given the relatively low K A /K S ratios of other branches of the tree. But the addition of crystallographic information, itself not compelling, makes a combination that is more compelling. Further, this hypothesis generation itself generates discoveries that might lead to their own hypotheses. An analysis of the evolutionary branches separating pigs and humans suggests an additional episode of adaptive change. The branch leading to the ancestor of human aromatase (branch 3) has a remarkably high K A /K S ratio (13 non-synonymous and no synonymous changes; Figure 5 ). This is a K A /K S ratio greater than unity, and does (pending evaluation of its statistical significance) compel the inference of an episode of adaptive change. Intriguingly, these changes are also clustered in the same regions of the structure as those changing along branch 1 leading to the stem fetal/placental enzyme, near the substrate and co-reductant binding sites. This implies that the substrate/product specificity of the ancestral aromatase protein was not like that of either the human or the pig placental forms, but rather reflects features that arose convergently in these two species [ 58 ]. Notably, four of the sites (positions 47, 153, 219, 269) that undergo replacement during the emergence of pig placental aromatase from the last common ancestor are the same as four that arose in the emergence of the human aromatase from its last common ancestor. Of these, the amino acid replacements are identical at two sites (Thr → Met at site 153; His → Arg at site 269). The probability associated with randomly observing this pattern is extremely low (0.000021) [ 59 ]. An additional site is displaced by a single position in the sequence alignment (259/260). We hypothesize that these represent an example of adaptive parallel evolution. It is important to point out that even an analysis this broad is likely to cover only a small part of a complicated reproductive endocrinology that must be associated with larger litter sizes. For example, the exact nature of the products produced by individual aromatases remains controversial, and may be different in laboratory studies depending on the conditions where they are studied [ 50 , 60 - 62 ]. This is especially the case with the 19-nortestosterone derivatives in Figure 1 . Further, an elegant recent study by Corbin et al. [ 23 ] identified 1β-hydroxytestosterone as a novel product produced by recombinant pig ovarian aromatase that is absent from the products produced by the porcine placental paralog, or by either human or bovine aromatase. This testosterone derivative binds to an androgen receptor, consistent with physiological activity. This was unknown before just this year, suggesting that more endocrine novelties remain to be discovered. Any of these may be relevant to a test of this system. For example, these hypotheses make predictions about the product specificities of the two peccary aromatases reported here. In fact, some data suggest that uterine exposure to androgens severely decreases litter size and embryonic survival during the time of maternal recognition of pregnancy [ 63 ]. This is consistent with the hypothesis of Corbin et al. [ 50 ] that the evolution of the placental paralog is associated with increased efficiency of testosterone aromatization. This is also consistent with the current data, and the argument presented here. It goes without saying that still more factors might be associated with an increase in litter size from one-two (presumed in Diacodexis , see Figure 4 ) to 12 or more in domestic swine. Most trivially, this increase might be associated with an increase in ovulation rate, and/or an adjustment in the structures and binding specificities of estrogen receptors [ 64 ]. Nevertheless, the first aromatase duplication, shared by pigs and peccaries, appears to have happened in the late Eocene (recognizing the error associated with these dates), around 35 Ma (Figure 4 ). This was a time of great global change, with dramatic cooling in the higher latitudes. More archaic kinds of mammals (e.g., some earlier families of perissodactyls and artiodactyls) became extinct, while many modern families (including the Suidae and Tayassuidae) became established at this time [ 65 ]. Suoids differed from other contemporaneous ungulates in their commitment to omnivory, even though a few forms, such as the modern warthog Phacochoerus aethiopicus , are more specialized herbivores. Perhaps the ability to bear a slightly larger litter than other artiodactyls was advantageous to them in this time of global ecological transition. However, it should be noted that larger litters usually mean altricial (i.e., relatively underdeveloped) young, a reproductive strategy apparently not available to larger, cursorial (running-adapted) ungulates, which give birth to precocial (i.e., well developed) young that are fully locomotory at birth [ 66 ]. The second aromatase duplication, with the ensuing capacity to produce multiple young, probably occurred within the family Suidae, some time during the Oligocene. The molecular data suggest dates of divergence between porcine fetal and placental aromatases as between 27–38 Ma, and the earliest known suid is of early Oligocene age [ 67 ], around 33 Ma (Figure 4 ). Large litters may have characterized the entire suid family. While the extant subfamily Suinae is primarily a Plio-Pleistocene radiation, during the Oligocene to Pliocene suids were exceedingly diverse taxonomically (with six other subfamilies known) as well as individually abundant as fossils [ 32 , 33 , 67 ]. In contrast, the predominantly North American tayassuids were never as diverse. It is possible that this tremendous Old-World diversity of suids, which continues to this day, is related to their capacity for the production of large litters. This type of speculation opens questions. For example, the babirusa (an Indonesian pig) is reported to have average litters of one-two individuals [ 68 , 69 ]. While it is possible that litters contain three-four individuals, the occurrence is low [ 70 ]. If the common ancestor of babirusa with the African/Eurasian Suinae had a larger litter, then the babirusa must be hypothesized to represent a reversion to the more primitive condition. At present, however, relatively little is known of either the molecular biology or the natural history of babirusa. The date of divergence from modern swine is placed between 12–26 million years [ 71 , 72 ], while our TREx analysis using cytochrome b places this data at ca. 18 Ma (data not shown). Clearly, further study is warranted. Conclusions The aromatase family offers an example where a combination of phylogenetic analysis, molecular evolutionary analysis, and chemical analysis set within the context of the paleontological and geological records, and supported by contemporary bioinformatics and molecular modeling tools, permits a higher order level of hypothesis generation concerning the function of proteins. Rather than simply an Enzyme Commission number (E.C. 1.14.14.1 for aromatase), a description of catalytic activity (the enzyme oxidizes testosterone), or a description of the regulatory pattern (the protein expressed between day 10 and 13), this type of analysis can generate a truly functional hypothesis: that the embryonic enzyme oxidizes testosterone as a way of managing the larger litter sizes that emerged in the Suoidea during a time of dramatic planetary cooling (ca. 35 Ma). Such hypotheses set a higher bar, and a more useful standard, for the field of systems biology. Evolutionary theory holds that the only mechanism for obtaining functional behavior in a biological system is natural selection. Selection, based on a frequently poorly defined concept of "fitness", is determined by a context that not only includes the cell and tissue, but also the organism, the ecosystem, and a changing planet [ 73 ]. One cannot expect a collection of expression data with a mathematical model, by themselves, to provide this type of functional information unless it is set in the organismic, ecosystem, and planetary context. The historical view, of the type outlined here, becomes a critical tool for constructing this setting (Supplementary Figure [see Additional File 1 ]). Humans have evidently exploited the molecular biology of larger litters to select for pigs that have truly large litters (as many as 14) following their domestication. Evidence for ancient domestication of pigs comes, inter alia , from a study of Indo-European languages. Proto-Indo-European (PIE) language had words for "pig" (PIE su -, compared with Tocharian B suwo , Latin sus , Greek us , Sanskrit sukara , Church Slavic svinija , Old High German swin , and English sow ; also compare PIE porko -, with Latin porcus , Church Slavic prase , Old High German farah , etc. [ 74 ]), indicating that the pig has been under human domestication for at least 6000 years, enough time to have suffered a significant impact on its genotype through husbandry. We are unable, at this time, to exploit complete genome sequences of pigs or other closely related taxa to discuss the impact of domestication on aromatase, steroid receptors, amphiregulins, or other proteins that appear to be associated with uterine capacity and large litter sizes in the domesticated pig [ 75 ]. With the anticipated complete genome sequences of representatives of various mammal orders, including artiodactyls, it should be possible to extend this planetary biology approach. Methods Calculations were done under the RedHat Linux 6.3 operating system on an Intel-Pentium III instrument using Blackdown's Java-SDK 1.1.8. PAML calculations were done on an IBM PC using the Unix operating system. Sequence analyses were aided by the DARWIN bioinformatics package [ 76 ]. The DARWIN package can be obtained by emailing a request to cbrg@inf.ethz.ch . Initially, pairwise alignments were constructed for the aromatase protein sequences available in the database. An evolutionary distance in PAM units was calculated for each pair by applying the PamEstimator-package from DARWIN using an empirical log-odds matrix. From this, a preliminary evolutionary tree was built for the mammalian sequences, with branch lengths along internal nodes calculated to minimize a least-squares distance. The sequences of the ancestral genes and proteins at branch points in the tree were then reconstructed. From there, mutations (including fractional mutations) at both the DNA level and protein level were assigned to individual branches in the tree using the method of Fitch [ 77 ]. The evolutionary history of the aromatase family was then analyzed using the transition redundant exchange (TREx) metric based on an analysis of two-fold redundant codon systems [ 24 , 78 ]. These were obtained for each pairwise comparison of aligned aromatase genes. The number ( n ) of two-fold redundant amino acids (Cys, Asp, Glu, Phe, His, Lys, Asn, Gln, and Tyr) that are conserved in the aligned pairs was determined. The number of those amino acids that are encoded by the same codon ( c ) was determined, and the fraction ( f 2 = c / n ) of the codons that are the same were then tabulated (Supplementary Table [see Additional File 2 ]). The TREx distances were calculated from f 2 values using the expression kt = -ln(( f 2 -E quil )/(1-E quil )), where E quil is the f 2 value expected after a large number of nucleotide substitutions have occurred at the synonymous sites [ 24 ]. The DNA sequences for aromatase were phylogenetically analyzed using a maximum likelihood framework in PAUP 4.0* (beta 10) [ 79 ], with the following parameters: alpha value representing the gamma distribution (2.1), the transition-transversion ratio (1.6), proportion of invariable sites (0.24), and empirical base frequencies. The resulting topology of the tree mirrors those based on other molecular studies [ 80 ]. For inter-taxon analyses, families in the MasterCatalog (EraGen Biosciences, Madison WI) were identified that contained at least one representative protein from both of the taxa of interest. For these families, all inter-taxa pairs of genes were extracted, together with the pairwise protein sequence alignment. A pairwise alignment of the DNA sequences was then generated to follow the protein sequence alignment. If a family contained more than one sequence of a species belonging to one of the taxa analyzed, then those sequences were checked to determine whether they were duplicate entries into the database. If this was the case, only one of the duplicate sequences was retained in the analysis. A histogram of inter-taxa pairs was constructed, and the f 2 value characteristic of orthologs determined [ 24 ]. This was used to calibrate the TREx clock using the divergence of pigs and oxen, and pigs and humans. Codon biases were obtained from the CUTG (Codon Usage Tabulated from GenBank) made available by the Kazusa DNA Research Institute Foundation, Japan [ 81 ]. Pairwise TREx distances were used to generate lengths for the branches connecting the swine paralogs using the minimum evolution criterion in PAUP. This preliminary analysis was followed by a maximum likelihood analysis for the complete dataset using the PAML program [ 82 ]. This includes the assignment of K A /K S values to individual branches. Tests of parallel evolution were conducted using Converge [ 59 ], implementing the JTT model. Secondary structural data based on homology modeling for aromatases were generated using the DARWIN bioinformatics package, and in agreement with previous studies [ 83 , 84 ]. Renderings of the three dimensional structure of the proteins were obtained using a beta version of the HyperProtein package (HyperCube, Gainesville FL, USA 32601). Authors' contributions EAG performed the evolutionary, statistical and structural analyses, and prepared the manuscript. LGG cloned genes as part of his Masters work, and called the evolutionary problem to the attention of SAB. TL provided computational infrastructure. RCMS and FAS initiated the work with suid reproductive endocrinology, and supervised LGG. DRS and DAL did the initial bioinformatics analysis. CMJ provided paleontological expertise, constructed the cladogram, and helped prepare the manuscript. SAB has developed planetary biological analysis as a paradigm for generating hypotheses about the biological function of proteins, and prepared the manuscript. Supplementary Material Additional File 1 Illustration of planetary biology. This figure illustrates the concepts of planetary biology as they relate to combining genomic, paleontological, chemical and ecological records to understand the history of the biosphere. Click here for file Additional File 2 An analysis of silent nucleotide substitutions in vertebrate aromatases. The first five columns from the left indicate the index number of sequence 1 compared with sequence 2, the fraction of sites at conserved two-fold redundant coding systems that are identical (f2), the number of such sites that are conserved (c2), and the number of such sites overall (n2). The remaining columns report analogous data: for silent sites in codon systems where a change at the third nucleotide is silent only if the change is a pyrimidine-pyrimidine transition (f2y, c2y, n2y); in silent sites where a change at the third nucleotide is silent only if the change is a purine-purine transition (f2r, c2r, n2r); for the silent sites at three-fold redundant codon systems (f3, c3, n3); and for the silent sites at four-fold redundant codon systems (f4, c4, n4). Click here for file	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC515309.xml
314462	Science on the Rise in Developing Countries	The disparity in the scientific output between developed and developing counties is dramatic, but, as the Americas show, this grim picture is improving	Kofi Annan, the Secretary-General of the United Nations, recently called attention to the clear inequalities in science between developing and developed countries and to the challenges of building bridges across these gaps that should bring the United Nations and the world scientific community closer to each other ( Annan 2003 ). Mr. Annan stressed the importance of reducing the inequalities in science between developed and developing countries, asserting that “This unbalanced distribution of scientific activity generates serious problems not only for the scientific community in the developing countries, but for development itself.” Indeed, Mr. Annan's sentiments have also been echoed recently by several scientists, who present overwhelming evidence for the disparity in scientific output between the developing and already developed countries ( Gibbs 1995 ; May 1997 ; Goldemberg 1998 ; Riddoch 2000 ). For example, recent United Nations Educational, Scientific, and Cultural Organization (UNESCO) estimates ( UNESCO 2001 ) indicate that, in 1997, the developed countries accounted for some 84% of the global investment in scientific research and development, had approximately 72% of the world researchers, and produced approximately 88% of all scientific and technical publications registered by the Science Citation Index (SCI). North America and Europe clearly dominate the number of scientific publications produced annually, with 36.6% and 37.5%, respectively, worldwide ( UNESCO 2001 ). North America and Europe clearly dominate the number of scientific publications produced annually. It is rather obvious that richer countries are able to invest more resources in science and therefore account for the largest number of publications. It is also likely that there is a statistical bias on the part of the SCI as a bibliometric database, since it represents North American and European publications far better than those of the rest of the world ( Gibbs 1995 ; May 1997 ; Alonso and Fernández-Juricic 2001 ; Vohora and Vohora 2001 ). But is the disparity in scientific contributions between the developed and developing worlds actually remaining unchanged or even increasing, as Mr. Annan has implied? A closer look at the trends over the last decade reveals important advances in developing countries. For example, Latin America and China, although representing, respectively, only 1.8% and 2% of scientific publications worldwide, have increased the number of their publications between 1990 and 1997 by 36% and 70%, respectively, which is a much higher percentage than the increments reached by Europe (10%) and industrial Asia (26%). The percentage of global scientific publications from North America actually decreased by 8% over the same period ( UNESCO 2001 ). Publishing Trends in the Americas Using the SCI databases produced by the Institute for Scientific Information (ISI), as well as data compiled by the Red Iberoamericana de Indicadores de Ciencia y Tecnología (RICYT), we examined the differences in the number and proportion of scientific publications between the developed world and the developing world from 1990 until 2000, focusing on the Americas as a case study. Not surprisingly, there was a huge disparity in the number of publications from 1990 until 2000, with the United States contributing the lion's share (84.2%), followed by Canada (10.35%). Latin America as a whole contributed only 5.45% to the total number of scientific publications in these ten years ( RICYT 2002 ). The total number of publications, however, is not necessarily the best measure for assessing scientific productivity or technical advances ( May 1997 ). More relevant measurements for these factors include the proportional change in the number of publications and the total number of publications when corrected for investment in research and development ( May 1997 ). The proportional change in the number of publications, using 1990 as a comparison, revealed that scientific publishing in Latin America increased the most rapidly in the Americas, far outpacing the United States and Canada ( Figure 1 ). Further analyses, correcting the number of overall publications for the amount of money invested in research and development for each region, also show that, in contrast to both Canada and United States, the trend in Latin America has been an increase in relative output throughout the 1990s ( Figure 2 ). Moreover, when taking into account the amount of research money available to researchers, Latin America actually out-published the United States and Canada by the year 2000 ( Figure 2 ). Although the cost of research is undoubtedly cheaper in the developing world due to relatively low researcher salaries, overhead and other work standards, these factors do not explain the substantial increase in the number of publications per amount of money allocated to research and development in Latin America, particularly from 1995 until 2000 ( Figure 2 ). Figure 1 Relative Increase in Scientific Publications in the Americas This figure shows the relative increase in publication in the Americas measured as the proportional change (%) in the number of SCI publications compared with the number of publications in 1990 ( RICYT 2002 ). Figure 2 Number of SCI Publications per Million Dollars This figure shows the number of SCI publications per million dollars that are invested in research and development in the Americas ( RICYT 2002 ). Other relative indicators of scientific productivity, such as the number of publications picked up by the SCI in relation to the number of scientists in a particular country, also demonstrate that such developing regions as Latin America are making substantial contributions to science, despite the fact that the average proportion of gross domestic product (GDP) invested in science in Latin America throughout this 10-year period was only 21% of the amount invested in United States ( RICYT 2002 ). Indeed, this scientific productivity is remarkable when we compare it with the relatively low investment in science itself as compared with the GDP of Latin America as a whole. In fact, Albornoz (2001) concluded that, as a group, Latin America could afford to invest a much higher proportion of its resources in scientific research and development. Latin American investment in research and development represented only 0.59% of the regional GDP in 1998, a very weak effort compared with that of the United States (2.84%) and Canada (1.5%). Among Latin American countries, there is a high degree of variability in publication rate as well as in financial investment in science and technology. Some countries have performed particularly well. For example, Uruguay, Chile, Panama, and Cuba averaged, respectively, 6.8, 5.3, 5.2, and 3.4 publications per million dollars of research and development investment in the 10 years studied, which is notoriously high compared with United States (1.5) and even Canada (3.3) ( RICYT 2002 ). Other countries, such as Costa Rica, Cuba, Brazil, and Chile, have invested a much greater proportion of their GDP in research and development than the other countries of this region ( Albornoz 2001 ). Why has the number of publications per dollar invested in research and development been increasing in Latin America while decreasing in United States and Canada? Explaining the Increase in Publishing Productivity in Latin America One potential explanation for the increase in scientific productivity in Latin America is that scientific development during the 1990s was particularly strong for many countries of this region. Indeed, this would explain the rapid rise in the number of publications in Latin America compared with the relatively flat increases in the United States and Canada, which were publishing just as well at the beginning of the decade. A potentially more important question, however, is why the number of publications per dollar invested in research and development has been increasing in Latin America while decreasing in the United States and Canada. This pattern could be the result of a variety of factors, none of which are mutually exclusive. It is possible that publishing in international journals as a measure of scientific productivity is becoming more important in Latin America. Increased funding to the most productive scientists from the national science development programs might have been an important stimulus. International cooperation resulting in more scientific collaborations among scientists in Latin America, Europe, and the United States may also have increased the relative number of publications in Latin America. In contrast, the decreasing trends in the number of publications per investment dollar in Canada and United States could reflect a trend towards more costly research in larger scientific programs. Scientific Impact from Latin America What, exactly, is the relative impact of such developing regions as Latin America on the scientific community? We used SCI 2001 data to examine the proportion of publications in the area of ecology (including the fields of evolutionary biology, conservation biology, and global change biology) between 1990 and 2002 in both the two top general science journals ( Nature and Science ; with impact factors of 27.96 and 23.33, respectively) and in the 20 top ecological journals (with impact factors of 10.51–2.47) ( ISI 2001a ). We credited a region with a publication if any of the authors were affiliated with institutions from that region. Thus, more than one region would receive credit for a single publication if that publication had been written by multiple authors from institutions of different regions. For the top 20 ecological journals, the American subcontinents of South, Central, and North America accounted for 62% of the publications worldwide. Within the Americas, however, Latin America represented only 6%, while Canada and United States accounted, respectively, for 13% and 82% of the top 20 ecological publications. When we examined the data as contributions to the top 10 ecological journals (impact factors 10.51–3.31) versus the top 11–20 (impact factors 3.28–2.47), the Latin American countries contributed nearly twice as many publications to journals in the second category (8% in the top 11–20 compared with 4% in the top 10). These findings suggest that publications from such developing regions as Latin America are falling short of reaching the top journals. In contrast, the United States contributed somewhat more publications to the top 10 journals (84%) than the top 11–20 journals (79%). The difference in the proportion of publications contributed by the United States to the top 10 and top 20 journals was even more pronounced when we examined it in respect to worldwide publications. In this case, the United States contributed 60% of the publications to the top 10 journals and only 40% of the publications to the top 11–20 journals. Interestingly, the proportion of publications from Latin America, the United States, and Canada across all subject areas in Science and Nature were nearly identical to those of the top 20 ecological journals. In Science and Nature , Latin America had 7% of the publications within the Americas versus 6% in the top 20 ecological journals, whereas the United States and Canada had 81% versus 82% and 12% versus 13%, respectively. These similarities suggest that the Latin American researchers are not shying away from the two top-ranked general science journals. However, publishing in Science and Nature was not enough to gain prominence, as evidenced by the number of citations of these researchers. The latest list of the 247 most-cited researchers in ecology and environmental sciences emphasizes the overwhelming contributions of authors from North America (73%) and Europe (21%) ( ISI 2001b ). No researcher working in a Latin American institution was included in the remaining 6%. Overall, these data indicate that the scientific output in the field of ecology in Latin America is having a relatively low impact in the international scientific community and is underrepresented in the top international journals, despite its robust productivity as measured by the number of publications per researcher funding amount. Similar findings were also reported for Asia ( Swinbanks et al. 1997 ) and thus could be a general phenomenon in the developing world. Although there are outstanding scientific researchers in the developing world who independently are making important contributions to the international scientific community, they are the exception. Why, in general, do Latin American scientists often fail to reach the top journals or become amongst the most cited researchers in their fields? One possibility is that the main research agendas between both regions are somewhat different and that the top journals, which are published in the developed world, respond more to the scientific mainstream of the developed regions. This is not to suggest any sort of conspiracy, but rather it implies that the perception of the most important science is linked to the region and that because the major funding agencies as well as most prominent journals share a similar economic region, they also share the same perception of what science is most interesting to them. Another consideration is that more local journals from developed regions are listed by the SCI than similar journals from developing regions ( Gibbs 1995 ). Consequently, there are more high-profile regional publication opportunities available to scientists from the developed region, whereas much of the research published locally in the developing world is overlooked. But it takes more than publishing good papers to become a highly cited scientist. It requires attending international meetings and introducing novel research findings in multiple scientific forums. Funding these activities, however, requires a greater proportion of research money being spent on meetings for researchers in the developing compared with the developed world. A Long Road Yet to Travel The positive trends in scientific productivity in Latin America should not be misinterpreted as a reason to be unconcerned about the existing gap highlighted by Mr. Annan. There are many compelling reasons for the push to increase scientific input from the developing world ( Goldemberg 1998 ; Annan 2003 ). One is that science, as a discipline, would benefit from the contributions of many disparate groups around the world, rather than being dominated by two geographic regions. Many scientific problems could be solved much more readily with the cooperation and scientific insight of scientists from developing regions. Climate change and biodiversity research, for example, urgently need the scientific input from those developing regions that are so important for these global processes. It is also critical for the developing world to promote, through research and publications, those areas of concern that are having a proportionally greater scientific and social impact upon them. There are now examples in which research on priority areas for the developing nations can actually become pioneering work in areas neglected by the research agenda of the industrialized world. This has been the case for research on renewable energy sources in Brazil ( Goldemberg 1998 ) and biomedical sciences in Cuba ( Castro Díaz-Balart 2002 ). These examples are important not only for those regions of the developing world, but are also in themselves scientific innovations that can greatly advance the knowledge of the rest of the world. Climate change and biodiversity research urgently need the scientific input from those developing regions that are so important for global processes. Although the evidence presented here demonstrates that there is a long way to go before developing countries contribute a more equitable share to the international scientific community, there are also reasons to be optimistic. The relative increase in the number of publications, especially when corrected for the amount of money available in research and development, demonstrates that many developing countries are heading in the right direction. The extremely high scientific productivity of many developing nations, corrected for and despite the rather limited availability of funds, suggests that increased funding to the sciences will be an excellent investment by developing nations in terms of publications as a measure of scientific output, particularly if these publications can target the journals that have the greatest impact. Although there may still be a long road to travel, we feel optimistic that the bridges mentioned by Mr. Annan are slowly being built.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC314462.xml
545489	A population-based study of anxiety as a precursor for depression in childhood and adolescence	Background Anxiety and depression co-occur in children and adolescents with anxiety commonly preceding depression. Although there is some evidence to suggest that the association between early anxiety and later depression is explained by a shared genetic aetiology, the contribution of environmental factors is less well examined and it is unknown whether anxiety itself is a phenotypic risk factor for later depression. These explanations of the association between early anxiety and later depression were evaluated. Methods Anxiety and depressive symptoms were assessed longitudinally in a U.K. population-based sample of 676 twins aged 5–17 at baseline. At baseline, anxiety and depression were assessed by parental questionnaire. Depression was assessed three years later by parental and adolescent questionnaire. Results Shared genetic effects between early anxiety and later depression were found. A model of a phenotypic risk effect from early anxiety on later depression provided a poor fit to the data. However, there were significant genetic effects specific to later depression, showing that early anxiety and later depression do not index entirely the same genetic risk. Conclusions Anxiety and depression are associated over time because they share a partly common genetic aetiology rather than because the anxiety phenotype leads to later depression.	Background Anxiety and depressive disorders are some of the most common psychiatric diagnoses in children and adolescents respectively. Both are known to be associated with major impairment in childhood and adverse consequences in later life [ 1 - 4 ]. Estimates of the prevalence of any childhood anxiety disorder are in the order of 3 to 12 % [ 1 , 4 ] and rise to as high as 40% or over if impairment is not required for a diagnosis [ 4 ]. In general, epidemiological studies show that rates of any anxiety disorder are higher in children than adolescents [ 5 ]. In contrast, rates of depressive disorder in young people show higher rates in adolescence (2 to 8 %) than in childhood (1 to 3 %) [ 6 ]. Depression and anxiety in childhood and adolescence have long-term deleterious outcomes for a significant proportion of young people. Depression and anxiety, once experienced in childhood are very likely to recur in adulthood [ 2 , 7 ]. Early onset depressive and anxiety disorders are also associated with substantial social impairment [ 8 ]. Even sub-clinical levels of depression in children and adolescents are associated with significant morbidity in the form of psychosocial impairment and service utilisation [ 9 , 10 ]. Furthermore, adolescents identified as having high levels of depressive or anxiety symptoms are significantly more likely to experience depressive disorder in adulthood than adolescents with depression levels within the normal range [ 11 , 12 ]. The observations that sub-clinical symptoms of depression are associated with significant morbidity, and that high levels of depression and anxiety symptoms predict depressive and anxiety disorders add to the evidence that depression and anxiety can be regarded as continua [ 13 - 15 ]. Depression and anxiety co-occur more commonly than would be expected by chance in children and adolescents. This co-occurrence has been identified both in clinical studies of children and adolescents and general population samples that have examined sub-clinical levels of depression and anxiety symptoms [ 16 , 17 ]. More specifically, anxiety symptoms or disorders most often precede depressive symptoms or disorders [ 18 - 21 ]. Moreover, although certain sub-types of anxiety, namely social phobia and panic rarely precede depression [ 22 ], individuals with these disorders and depression are very likely to have had a different anxiety disorder that predated the onset of depression [ 20 ]. Twin studies provide a means of examining the extent to which the genetic and environmental aetiological factors contributing to two disorders or symptom groups overlap and to what extent they are distinct. Longitudinal data, collected at more than one time point provide a further test, namely, that one set of symptoms or disorder is a risk factor for another. This approach is especially useful in the study of anxiety and depression. Several groups have suggested that anxiety may be a developmental precursor of depression, particularly in young people who are at increased risk of depression due to parental depression [ 23 - 25 ]. Indeed Kovacs & Devlin [ 17 ] suggested that children may be biologically 'prepared' to experience symptoms of anxiety rather than depression. The results of other studies are consistent with this proposal. For example, in a sample of depressed children, in those children who had a comorbid anxiety disorder, the anxiety disorder was found to have preceded depressive disorder in two thirds of cases [ 18 ]. Similar evidence that anxiety disorders tend to precede depression has been reported in longitudinal epidemiological [ 19 , 21 , 26 ] and clinical studies [ 20 ]. Indeed, a recently convened National Institute for Mental Health (NIMH) workgroup recommended research into childhood anxiety as a known precursor of depression as a priority [ 27 ]. Despite clear indications that anxiety and depression in childhood and adolescence are associated, it remains unclear as to how the transition from anxiety to depression is mediated over time. Possible factors include aetiological factors in common – these may be 1) genetic or 2) psycho-social risk factors, or 3) a direct risk effect of anxiety leading to later depression. Cross-sectional twin studies of children [ 28 , 29 ] and adults [ 30 ] and one longitudinal twin study of girls [ 31 ] have shown that to a large extent, the overlap between anxiety and depression is due to a common set of genes that influence both depression and anxiety. However, shared environmental factors have also been shown to be important sources of covariation between anxiety and depression symptoms for children but not adults [ 28 , 31 ] which suggests the importance of shared psycho-social risk factors for anxiety and depression. Nevertheless, two out of these three twin studies were based on cross-sectional data and were therefore not able to determine the genetic and environmental associations between anxiety and depression over time. Furthermore, despite the importance of understanding why anxiety tends to precede depression, [ 27 ] no twin study of children and adolescents has yet specifically tested the hypothesis that anxiety is a phenotypic risk factor for depression. The present study also adds to the existing literature in that data on depression symptoms from different raters (mother and child) are available, thus allowing associations to be examined with data from different informants. In the present study, we set out to examine two hypotheses that may explain the observed associations between early anxiety and later depression. 1. Early anxiety symptoms and later depression symptoms are associated because of shared risk factors. 2. The association between anxiety symptoms and later depression symptoms is mediated by a risk effect of the phenotype of anxiety. Method Participants Families from a systematically ascertained, population-based register of all twin births between 1980 and 1991 in South Wales, U.K. were invited to participate. This register forms a sub-sample of the Cardiff Study of All-Wales and North West of England Twins (CASTANET). Twins who had emigrated were excluded, as were cases in which one of the twins had died or had a serious illness. At the first wave of data collection in 1997, there were a total of 1109 pairs of twins aged 5–17 years although not all of these individuals were eligible to participate at both time points (see below). Data were collected by postal questionnaire. Families received three reminders and telephone reminders when numbers could be traced. The same methods were used three years later to collect longitudinal data except that families received four reminders. To be invited to participate in the follow-up study, we required that the twins were living together in the same home and were under the age of 18 years. Twins were required to live in the same home in order to minimise heterogeneity of environmental risk factors that can impact on genetic and environmental parameter estimates. The focus of the follow-up study was childhood psychopathology and for that reason young people aged 18 and over were not included. At time 1, there were 986 twin pairs who were within the age range of the study at both time points. In the first wave of the study (1997; time 1), 670 families provided questionnaire responses giving a response rate of 61%. Comparison of responders and non-responders using Townsend Scores [ 32 ] which index the level of deprivation of an electoral area revealed no significant socio-demographic differences between the two groups at time 1 (t = .373, p = 0.709). Families with children aged 8–17 were re-contacted in 2000 (time 2). Of the 670 families who replied at time one, 85 had moved away, there were 8 new contraindications and there were 123 children who were out of the age range of the study and did not live in the same home. This left a total of 454 families who were eligible at time 2. Of these, 338 families replied, giving a total response rate of 75%. There were no significant socio-demographic differences between responders and non-responders at time 2 (t = 1.71, p = 0.09). Zygosity was assigned using a twin similarity questionnaire which has been shown to be over 90% accurate in distinguishing identical (monozygotic; MZ) from fraternal (dizygotic; DZ) twins [ 33 ]. There were 198 MZ girls (99 pairs), 134 MZ boys, 128 DZ girls, 116 DZ boys, 270 opposite sex DZ twins. Measures At time 1, parents were asked to complete the Children's Revised Manifest Anxiety Scale [ 34 ] which assesses symptoms over the past three months. It has previously been found to be a reliable and valid instrument [ 35 ] (Cronbach's α = .8662 twin 1, α = .8708 twin 2). Parents also completed the general functioning scale of the McMasters Family Assessment Device (FAD) [ 36 ]. At both time points, parents completed the short version of the Mood and Feelings Questionnaire (MFQ) [ 37 ]. At the second wave children aged 11 or above also completed the MFQ. The MFQ is based on DSM-III-R symptoms of depression and has been successfully used as a screening questionnaire for clinical depression in community populations [ 38 ] (α = .9231 twin 1, α = 9320 twin 2). Analysis Descriptive statistics For descriptive statistics, (correlations and mean comparisons), the survey commands in the program STATA [ 39 ] were used. These commands take into account the clustering of the data from twin pairs (i.e. each twin pair provides two data points) by likening the twin data to a two-stage cluster design with the twin pairs as the primary sampling unit. Since reliability coefficients can not be calculated using these commands these were presented for first and second-born twins separately. Univariate Analysing data from twins provides a means of estimating the relative contribution of genetic and environmental effects on individual variation in behaviour. In the basic (ACE) model, variation can arise from three sources: 1) additive genetic effects (A); 2) common environmental effects (C); 3) unique environmental effects (E). Common environmental effects are non-genetic factors that serve to make twins more similar to one another while unique environmental effects are non-genetic factors that uniquely influence one individual within a twin pair and tend to make the individuals in a twin pair different from each other. Model fitting was carried out using the programs Mx [ 41 ] and LISREL [ 42 ] and continuous measures of anxiety and depressive symptoms were analysed. The significance of the A, C and E parameters can be tested by dropping them from the model and comparing the fit of the reduced model to that of the full model using the χ 2 critical value for the number of degrees of freedom gained in the reduced model. Bivariate Bivariate analysis allows the covariance of two traits to be partitioned into covariance that is due to additive genetic factors, common environmental factors and unique environmental factors. The covariance parameters for the Cholesky model presented (see figure 1 ) include those factors in trait 1 (anxiety) that also influence trait 2 (depression). A bivariate model in which anxiety symptoms at time 1 precede depressive symptoms at time 2 was fitted consistent with clinical and epidemiological data showing that anxiety precedes depression more often than vice versa. In addition, a 3 variable model that included anxiety and depressive symptoms at time 1 and depressive symptoms at time 2 was fitted. This model estimated the genetic and environmental associations between anxiety at time 1 and depression at time 2 when the effects of concurrent depression were included. A causal model was then fitted (see figure 2 ). Comparing the fit of this causal model to that of the general bivariate (Cholesky) model allows two competing explanations of the association between anxiety (time 1) and depression (time 2) to be tested: 1) the association of anxiety and depression is due to genetic and /or environmental risk factors common to both anxiety and depression: 2) the association is due to a risk effect of the phenotype of early anxiety on later depression. A unidirectional causal model from anxiety to depression was fitted given that the data presented are longitudinal. Although the reliabilities of the anxiety and depression scales were good and comparable, a causal model that included residual error was included in line with the suggestion of Neale & Cardon [ 43 ]. This was fitted since in direction of causation models it cannot be assumed that measurement error will be confounded with non-shared environmental effects [ 44 ]. Fitting this type of model does not constrain measurement error to be transmitted phenotypically and thus is likely to provide more realistic parameter estimates than a casual model without residual error terms. Figure 1 Bivariate Cholesky decomposition of anxiety at time 1 and depression at time 2. Aanx genetic influences on anxiety Canx common environmental influences on anxiety Eanx non shared environmental influences on anxiety Ac genetic influences on anxiety that also influence depression Cc common environmental influences on anxiety that also influence depression Ec non shared environmental influences on anxiety that also influence depression Adep genetic influences specific to depression Cdep common environmental influences specific to depression Edep non shared environmental influences specific to depression Figure 2 Unidirectional causal model from anxiety at time 1 to depression at time 2 Aanx genetic influences on anxiety Canx common environmental influences on anxiety Eanx non shared environmental influences on anxiety Adep genetic influences specific to depression Cdep common environmental influences specific to depression Edep non shared environmental influences specific to depression Sex effects Univariate analyses were performed to test for both quantitative and qualitative sex differences. Quantitative sex differences, are tested by estimating the size of parameter estimates for the genders ('the common effects sex limitation model'). Qualitative sex differences test whether the set of genes influencing the phenotype differs by gender i.e. different genes ('the general effects sex limitation model'), this is done by estimating the genetic correlation for opposite sex DZ pairs and comparing the fit of this model to one that constrained the genetic correlation to 0.5. In addition, a bivariate sex limitation model was tested [ 45 ]. This model estimates whether the covariation between anxiety and depression is different for boys and girls. All analyses reported were based on the five twin groups (MZ male, MZ female, DZ male, DZ female, DZ opposite sex). Results Descriptive statistics There were no significant mean differences on anxiety or depressive symptoms by gender (t = -0.047, p = .962; t = -1.628, p = .104). Age was not associated with anxiety or depressive symptoms (r = .014, p = .707; r = .082, p = .078). The mean age of children at time 1 was 10.58, range 5.58–17.83, and at time 2, was, 12.64, range 8.75–17.25. Symptoms of early anxiety and later depression were strongly correlated (parent-rated symptoms r = .479. The correlation was slightly lower for symptoms across rater (parent rated anxiety and adolescent rated depression), r = .335. Sex effects For anxiety, univariate analysis of parent-rated data showed no significant gender differences in the magnitude of genetic parameter estimates (Δ χ 2 = 1.505, Δ df = 3), nor were there qualitative genetic differences (Δ χ 2 = 0, Δ df = 1). For depression, univariate models for parent-rated and self-rated scores indicated no significant gender effects for the magnitude of genetic effects (parent rated, Δ χ 2 = 2.679, Δ df = 3; self-rated, Δ χ 2 = 0.586, Δ df = 3) nor qualitative gender differences (parent rated, Δ χ 2 = 0, Δ df = 1; self-rated, Δ χ 2 = 0, Δ df = 1). Finally, for parent rated symptoms, results from the bivariate sex limitation Cholesky model showed no significant gender differences in the covariation between anxiety and depression in that the genetic and environmental covariation could be equated across the genders with no significant deterioration in fit (parent-rated, Δ χ 2 = 0.445, Δ df = 3). However, for self-rated symptoms, one parameter, i.e., the non-shared environmental covariation parameter, could not be equated across the genders (Δ χ 2 = 6.107, Δ df = 1). Estimates from this model for boys were; Aanx = 50, Canx = 17, Eanx = 32, Ac = 10, Cc = 47, Ec = -11, Adep = 9, Cdep = 5, Edep = 40; and for girls were; Aanx = 47, Canx = 22, Eanx = 30, Ac = 12, Cc = 13, Ec = 8, Adep = 30, Cdep = 12, Edep = 24; χ 2 = 30.93, df = 32, AIC = -33.07). These results are presented in addition to those for the combined sample (see table 1 ). Table 1 Bivariate model fitting for time 1 anxiety and time 2 depression Rater Aanx Canx Eanx Ac Cc Ec Adep Cdep Edep χ 2 AIC Parent rated total sample NMZ = 138 NDZ = 210 46* 24* 30* 13 36*** 2 ns 24* 0 ns 26* 15.36 df = 11 -6.64 Parent rated – adolescents only (8–14 at time 1 and 11–17 at time 2) NMZ = 90 NDZ = 136 46* 23 31* 11* 34** 4 ns 25* 0 ns 25* 15.02 df = 11 -6.98 Parent rated anxiety time 1 and self rated depression time 2 NMZ = 92 NDZ = 128 53* 15 ns 32*** 13* 17 ns 2 ns 36 0 ns 31* 5.80 df = 11 -16.20 Ns = non significant * p = .05 p = .01 * p = .001 Aanx genetic influences on anxiety Canx common environmental influences on anxiety Eanx non shared environmental influences on anxiety Ac genetic influences on anxiety that also influence depression Cc common environmental influences on anxiety that also influence depression Ec non shared environmental influences on anxiety that also influence depression Adep genetic influences specific to depression Cdep common environmental influences specific to depression Edep non shared environmental influences specific to depression NMZ = number of MZ twin pairs NDZ = number of DZ twin pairs Parameter estimates equated to sum 100 in direction of arrows on each trait (anxiety time 1 & depression time 2) Bivariate analysis Table 1 shows results from bivariate analyses of anxiety and depression. Both the genetic covariation (Ac) between anxiety and depression as well as the common environmental covariation (Cc) were significant, while unique environmental covariation (Ec) was negligible. However, despite significant covariation, there were also significant genetic (Adep) and unique environmental effects (Edep) specific to later depression. Thus, although significant genetic covariation between anxiety and depression was observed, the genetic effects on depression were not entirely mediated through genetic effects that were common with anxiety. This illustrates that the genetic covariation between anxiety and depression over time is not complete. Moreover, this observed genetic covariation does not appear to derive from the association of early depression with later depression in that the genetic covariation between anxiety and later depression remained significant when early depressive symptoms were included in the model (see figure 3 ). Figure 3 shows that there is a significant genetic path linking early anxiety and later depression (Ac (anx1 dep2) = 11, p = .001). Figure 3 Trivariate Cholesky decomposition of anxiety at time 1, depression at time 1 and depression at time 2 χ 2 = 39.74, df = 24, AIC = -8.26 Ns = non significant * p = .05 p = .01 * p = .001 Parameter estimates equated to sum 100 in direction of arrows on each trait (anxiety time 1, depression time 1 & depression time 2) It has been shown that the aetiology of depression varies significantly according to age, with greater common environmental effects in children aged 8–10 than in adolescents aged 11–17 [ 28 , 46 , 47 ]. The three year follow-up period of the study meant that we were unable to analyse children's and adolescent's symptoms separately. However, given previous findings of age effects, we carried out analyses including only those twins who were 'adolescents', that is, aged 11 or over at the second time point with the expectation that the common environmental covariation path (Cc) might decrease. Nonetheless, a significant common environmental component of variance remained. Family functioning and rater effects Following this finding of significant common environmental covariation the nature of this latent factor was further examined. Information on a potential common environmental variable, general family functioning (all twins lived in the parental home) was available. Family functioning was correlated with anxiety symptoms at time 1 (r = .210, p = .001). Bivariate analyses controlling for family functioning are shown in table 2 and it can be seen that this exerted only a slight effect on the estimate of common environmental covariation (drop of Cc from .36 to .31). However, associations between anxiety and depression symptoms could be due to the fact that a single informant (parents) rated their children's anxiety and depression symptoms at both time points. In bivariate analyses, similarities between variables that are due to shared rater effects will be partitioned into the common environmental component of covariance. A distinction should be made here between shared rater effects and rater bias. Rater biases that result in common environmental effects arise when a proxy informant, usually a parent rates a pair of twins as more similar than more objective measures would find them. This sort of rater bias results in deflated MZ phenotypic variance compared to DZ variance [ 43 ]. This pattern of variance has not been observed in this sample (anxiety DZ standard error = .035, MZ standard error = .048, t = 1.89, p = .06; depression DZ standard error = .056, MZ standard error = .067, t = .514, p =.608), in fact the MZ variance is greater than the DZ variance. Thus, rater bias can not account for the observed common environmental effects. On the other hand, shared rater effects come about simply as an effect of the same informant rating two sets of symptoms or risk variable and outcome and are therefore not exclusive to parental ratings. In order to test any potential shared rater effects, a bivariate model with data from different informants was tested (parent-rated anxiety symptoms at time 1 and adolescent-rated depression symptoms at time 2). It can be seen from Table 1 that the common environmental covariance influencing anxiety and depression (Cc) is no longer significant when cross-informant information is used. This suggests that at least a proportion of the Cc estimate observed in analyses that used only parent-rated data is likely to be due to shared rater effects, i.e. that part of the shared environmental covariation is due to the fact that the same informant rated both phenotypes. The observation that when family functioning was included as a measured environmental variable in the analyses of parent-rated information, the common environmental covariance estimate was only slightly reduced is consistent with the possibility that these Cc effects may be due to shared rater effects. However, given the small effect sizes of most single measured risk factors (genetic or environmental) [ 48 ], one might not expect single environmental risk factors to account for large proportions of variance. Indeed, risk variables for symptoms of depression and anxiety are likely to have multiplicative effects [ 48 , 49 ]. With this in mind, the same family functioning was also included in a cross-informant model. As can be seen from table 2 , including family functioning resulted in a small decrease in the Cc estimate (17 to 14). Table 2 Bivariate model fitting for time 1 anxiety and time 2 depression controlling for measured environmental variables Rater Aanx Canx Eanx Ac Cc Ec Adep Cdep Edep χ 2 AIC Parent rated total sample - family functioning at time 1 regressed out 47* 21 32* 12 31** 1 ns 24* 0 ns 26* 12.59 df = 11 -9.41 Parent rated anxiety and self rated depression - family functioning at time 1 regressed out 56* 11 ns 33*** 14* 14 ns 2 ns 38 0 ns 33* 6.75 df = 11 -15.25 Ns = non significant * p = .05 p = .01 * p = .001 Aanx genetic influences on anxiety Canx common environmental influences on anxiety Eanx non shared environmental influences on anxiety Ac genetic influences on anxiety that also influence depression Cc common environmental influences on anxiety that also influence depression Ec non shared environmental influences on anxiety that also influence depression Adep genetic influences specific to depression Cdep common environmental influences specific to depression Edep non shared environmental influences specific to depression NMZ = number of MZ twin pairs NDZ = number of DZ twin pairs Parameter estimates equated to sum 100 in direction of arrows on each trait (anxiety time 1 & depression time 2) Phenotypic "causal" model Finally, a model that included a direct causal path from anxiety to depression was fitted (see figure 2 ). This model allows for testing whether the phenotype of anxiety (rather than shared genetic and environmental aetiological factors) was responsible for the observed covariance between anxiety and later depression symptoms. That is, testing whether anxiety (independent of shared genetic and environmental risk factors) is itself an early risk factor for depression. The full causal model provided a significantly poorer fit than the general bivariate model (Δ χ 2 = 48.82, Δ df = 1, p < .001). Thus, it does not appear that anxiety leads to depression through direct phenotypic effects, but that, anxiety and depression symptoms are associated over time because they share aetiological factors in common. Discussion This investigation has used a longitudinal, epidemiological and genetically sensitive design to examine two possible explanations of the association between early anxiety and later depression symptoms in children and adolescents: 1) a common genetic/environmental aetiology or 2) a phenotypic risk effect of early anxiety. There was significant genetic covariation between anxiety and later depression. (Moreover, the genetic covariation was not explained by the effects of early depression). This result is consistent with the association between early anxiety and later depression being due to a common genetic aetiology. However, the genetic overlap between early anxiety and later depression was far from complete in that there were significant, separate genetic effects on anxiety and genetic influences specific to later depression. Thus, in this sample, symptoms of anxiety and depression in children and adolescents share only a partly common genetic aetiology. In addition to genetic covariation, significant shared environmental covariation between anxiety and depression was also observed. It appeared that some of the shared environmental covariation between anxiety and depression observed in parent ratings of anxiety and depression was due to shared rater effects as the common environmental covariation (Cc) was no longer significant when analyses were performed with data across different informants. However, including family adversity as a measured environmental risk factor into a model with data from different informants also resulted in a small decrease in the Cc parameter estimate. The model that included a direct causal path from early anxiety symptoms to later depression symptoms provided a significantly poorer fit than the general bivariate model. These results suggest that anxiety is not an aetiologically distinct phenotype that is in itself a risk factor for future depression symptoms, but rather that the covariation over time arises from the common genetic and environmental architecture. It should be noted though, that some of the common environmental covariation is likely due to shared rater effects, because we found that this path became non significant in cross-informant analyses. The present findings are consistent with those of several other twin studies that have reported strong genetic correlations between symptoms of anxiety and depression in children and adolescents [ 28 , 29 , 31 ] and adults [ 30 ] and with a study that found significant genetic effects specific to depression [ 29 ]. However, only one of these studies was longitudinal [ 31 ] and this included girls only, and none of these studies included information from more than one informant. Sex effects The lack of significant univariate and bivariate gender differences in genetic and environmental parameters estimates for parent reports in the present sample is of interest. The results for self reports of depression are less clear, previous analysis of a larger sample, from which the present sample was drawn, did find sex differences for self rated depressive symptoms as measured by the long version of the Mood and Feelings Questionnaire (MFQ) [ 47 ], which were not detected in the present sample. However, a previous cross-sectional analysis of the full time 1 self-rated depression data did not find significant gender differences [ 54 ]. The only significant gender difference in the present analysis was for the non-shared environmental covariation between anxiety and self-rated depression. The lack of significant effects for univariate analysis for self-rated depression may be due to the smaller sample size and thus lower power to detect effects in the present sample, or it could be due to the fact that different versions of the MFQ that were used in the present (short version) and a previous analysis (long version) [ 47 ]. Moreover, the bivariate Cholesky sex limitation analysis for self-reported depression was likely under-powered as few of the parameter estimates reached statistical significance. The sample size is small for those who provided self-reports (NMZ = 92 and NDZ = 128, see table 1 ) and it is therefore uncertain how reliable these results are. The non-shared environmental covariation estimate for boys also yields a negative parameter estimate (-.11) (albeit non-significant) which indicates the non-shared environment for anxiety is negatively correlated with the non-shared environment for depression. This finding is difficult to interpret, further suggesting caution in conferring too much confidence to the gender-specific findings in this model. Given the sample size for cross informant models in this study, it is not safe to draw firm conclusions about gender differences in the covariance of anxiety and depression when depression is self rated. This needs to be examined in a larger sample. However, although the prevalence of depression shows gender differences in adolescence, this does not necessarily suggest gender differences in aetiology. How do the present findings fit with results from family studies? Several family studies of the offspring of depressed parents have found increased rates of anxiety rather than depressive disorders [ 23 , 25 ] and Rende and colleagues [ 24 ] found that sibling resemblance for anxiety disorders was increased in the offspring of depressed parents. This familial aggregation of anxiety disorders could be due to common environmental or genetic factors. There is now consistent evidence from cross-sectional and longitudinal twin studies of children and adolescents that this observed familial association between anxiety and depression symptoms has a partly common genetic aetiology. Limitations As mentioned previously, several groups have shown that the aetiology of depressive symptoms differs between children (8–10) and adolescents (11–17) [ 28 , 46 , 47 ]. The majority, though not all, of the present sample were 'children' aged under 10 (range 5–14) at time 1 and 'adolescents' aged 11 and above (range 8–17) at time 2. Thus, as there is age heterogeneity in aetiology, high levels of effects specific to each time point might be expected such as shown in the present study. Nonetheless, we might not expect to find complete genetic overlap between anxiety and depression. For instance, the genetic liability to depression and antisocial behaviour in children and adolescents has also been shown to overlap [ 50 ]. Thus, there may be different developmental pathways to depressive symptoms in adolescence. In addition, previous studies have suggested that gene-environment correlation [ 51 , 52 ] and gene-environment interaction [ 49 , 53 ] involving life events also contribute to genetic variance in adolescent depression and such effects would also be subsumed within the estimate of genetic variance. Clinical implications Anxiety and depressive symptoms are strongly associated over time. This association does not appear to be due to a phenotypic risk effect of early anxiety. Rather, early anxiety and later depression are associated due to a common aetiology. This was primarily a common genetic aetiology although family functioning and a single informant rating on both sets of symptoms also contributed to this association. Some of the common genetic aetiology may act as indirect genetic effects via behaviour (gene-environment correlation and gene-environment interaction). Competing interests The author(s) declare that they have no competing interests. Author contributions AT and FR conceived the paper. FR carried out statistical analysis and wrote the paper. AT and MBM wrote and edited the paper. MBM provided statistical support. All authors read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC545489.xml
515296	A supertree approach to shorebird phylogeny	Background Order Charadriiformes (shorebirds) is an ideal model group in which to study a wide range of behavioural, ecological and macroevolutionary processes across species. However, comparative studies depend on phylogeny to control for the effects of shared evolutionary history. Although numerous hypotheses have been presented for subsets of the Charadriiformes none to date include all recognised species. Here we use the matrix representation with parsimony method to produce the first fully inclusive supertree of Charadriiformes. We also provide preliminary estimates of ages for all nodes in the tree. Results Three main lineages are revealed: i) the plovers and allies; ii) the gulls and allies; and iii) the sandpipers and allies. The relative position of these clades is unresolved in the strict consensus tree but a 50% majority-rule consensus tree indicates that the sandpiper clade is sister group to the gulls and allies whilst the plover group is placed at the base of the tree. The overall topology is highly consistent with recent molecular hypotheses of shorebird phylogeny. Conclusion The supertree hypothesis presented herein is (to our knowledge) the only complete phylogenetic hypothesis of all extant shorebirds. Despite concerns over the robustness of supertrees (see Discussion), we believe that it provides a valuable framework for testing numerous evolutionary hypotheses relating to the diversity of behaviour, ecology and life-history of the Charadriiformes.	Background The shorebirds and allies (Aves: Charadriiformes; [ 1 ]) present an exceptional group for studying numerous evolutionary hypotheses. Their remarkable diversity of social mating system, parental care, sexual dimorphism, ecology and life-history make them an ideal group for unravelling the mechanisms of, for example, sexual selection and sexual conflict. Previous comparative studies have made significant contributions to our understanding of the evolution of mating systems [ 2 ], parental care [ 3 , 4 ], sexual size dimorphism [ 5 - 7 ], locomotion and morphology [ 8 ], migratory behaviour [ 9 ], egg size [ 10 ], and plumage colouration [ 11 ]. The importance of phylogeny in cross-species comparative studies is well documented [ 12 - 14 ]. Large and well-resolved phylogenies that incorporate divergence times provide powerful tests of a wide range of hypotheses whilst accounting for the effects of shared evolutionary history [ 13 , 15 ]. However, the shorebird studies listed above were limited by the lack of a complete phylogeny for the group. Most of these studies are based on derivations of the seminal work of Sibley and Ahlquist [ 16 ], yet this study included less than a quarter of extant and recently extinct shorebird species. Recently extinct taxa (according to Monroe and Sibley [ 1 ]) are: the Tahitian sandpiper Prosobonia leucoptera , the Canary Islands oystercatcher Haematopus maedewaldoi , and the Great auk Pinguinus impennis . Recent molecular studies covering a wide range of shorebird families have drawn attention to conflict in the reconstruction of the deep basal nodes of shorebird phylogeny (figure 1 ; reviewed by van Tuinen et al. [ 17 ]). For example, morphological data [ 18 , 19 ] places Alcinae (auks, puffins, murres) at the base of the shorebird tree whilst sequence [ 20 - 22 ] and DNA-DNA hybridisation [ 16 ] data suggests that they are a highly derived sister group to Stercorariini (skuas and jaegers), Larini (gulls), Sternini (terns), and Rynchopini (skimmers). It is important to note that taxon coverage differs between these studies and this may be an important factor in determining the tree topology. Specific phylogenies have been derived, for example, for sandpipers [ 23 ], the genus Charadrius [ 24 ], and jacanas [ 25 ] using DNA sequence data. In contrast, morphological evidence provided the basis for Chu's [ 26 ] study of gull phylogeny. Strauch [ 18 ] presented the most complete data set of 227 Charadriiformes species. However, despite the plethora of cladograms for particular shorebird groups (see reviews by Sibley and Ahlquist [ 16 ]; Thomas et al. [ 22 ]), those that address relationships across the whole clade use either sparse taxon sampling [ 16 , 27 ], or are based on reassessments of Strauch's [ 18 ] data [ 19 , 28 - 30 ]. Note that Dove [ 30 ] included a feather microstructural analysis in addition to her reanalysis of Strauch's [ 18 ] data. Figure 1 Previous hypotheses shorebird phylogeny. Family and subfamily level relationships of shorebirds based on: a) Morphological data [19]; b) DNA-DNA hybridisation [16]; c) Sequence analysis of RAG-1 [20, 21], cytochrome- b [22] and myoglobin intron II [21]. Combining phylogenetic data Numerous methods and types of data can be used to infer phylogeny. Frequently, as in Charadriiformes, a single analysis incorporating all taxa of interest is absent. Under the principle of total evidence [ 31 ], all sources of phylogenetic information should be combined to maximize their explanatory power. Eernisse and Kluge [ 32 ] define total evidence as a method for seeking the best fitting phylogenetic hypothesis for an unpartitioned set of synapomorphies (shared derived characters) using character congruence (characters combined in a supermatrix). Hence, this method combines the primary data (molecular, morphological and behavioural characters) into a single analysis. The approach is powerful because weak signals in the partitioned data sets may be enhanced when combined, and previously obscured relationships may be revealed [ 33 ]. The total evidence approach has both practical and theoretical problems. First, only certain types of data can be combined. For example, nucleotide sequences and morphological traits can be readily assessed together as characters, but it is not generally possible to include nucleotide sequences and genetic distance data in a single analysis [ 34 ]. We acknowledge that Lapointe et al. [ 35 ] suggest a distance based approach to combine otherwise incompatible data in a total evidence analysis, although this method has not been tested beyond a single application. The consequence is that it is rarely possible to combine all sources of data in practice and the lack of overlap in combinable data sets may result in a reduction of the number of taxa included. Second, Miyamoto and Fitch [ 36 ] contend that combining data sets is rarely justified because partitions of phylogenetic data are real and unequivocal. They argue that several partitions producing similar topologies provide multiple lines of independent evidence supporting that topology. Theoretical arguments over the benefits of total evidence will undoubtedly continue, but perhaps the major barriers to its use are the often very high computational demands of large matrices, and the a priori exclusion of certain data types. This is particularly true of Charadriiformes phylogeny, where one of the most significant contributions to the field – DNA-DNA hybridisation – cannot be included. An alternative set of techniques, collectively termed supertrees (e.g., Matrix Representation with Parsimony, MRP; [ 37 , 38 ]), enables combination of trees (rather than raw data) from otherwise incompatible sources. MRP methods code source phylogenies based on the presence and absence of taxa at each node of the tree [ 37 - 39 ] and are thus one step removed from the primary data. It is important to recognise that supertrees should not be regarded as a replacement for exhaustive phylogenetic studies of the primary data and there are drawbacks to the methods (see Discussion). However, they do enable very large phylogenies to be constructed rapidly [ 15 ]. Supertrees have been constructed successfully for a wide variety of taxa including carnivores [ 15 ], primates [ 39 ], seabirds [ 40 ], dinosaurs [ 41 ], and grasses [ 42 ]. Shorebirds are particularly well suited for supertree treatment, since there are numerous incomplete phylogenies available and a broader phylogeny is desirable to facilitate powerful analyses of numerous evolutionary hypotheses (see above). Here, we present the first complete composite phylogeny of extant and recently extinct [ 1 ] shorebirds using the MRP approach. We are therefore combining data on tree topologies, and not conducting a simultaneous analysis on the original data. We also use fossil and molecular data to estimate divergence times (see Methods). The combination of complete taxonomic coverage and the inclusion of branch lengths provide the basis for future comparative analyses of Charadriiformes evolution. In addition, conflicting and unresolved areas of Charadriiformes phylogeny are revealed. Results and Discussion Supertree resolution and topology We found 1469 equally short trees of length 1847 steps using the parsimony ratchet approach (see Methods). This compares favourably to a standard heuristic search that yielded shortest trees of 1853 steps. All subsequent results and discussion refer to the parsimony ratchet analyses. Figure 2 shows the family and subfamily level relationships of shorebirds based on the strict and 50 % majority-rule consensus tree (see additional file 1 for branch length estimates). Figures 3 , 4 , 5 , 6 , 7 , 8 , 9 show the species level phylogeny. The full 50% majority rule consensus and the strict consensus trees are available as additional file 2 and 3 respectively. The 50% majority-rule consensus tree is well resolved (73.1%; 255 nodes out of a possible 349 in a fully bifurcating tree), although the strict consensus tree is only 49.6% resolved (173 from 349 possible nodes). The majority rule tree includes nine novel clades (numbers 20, 29, 57, 85, 89, 108, 122, 139, 140) that do not appear in any of the source trees; all of these occur towards the tips of the tree. This is a general problem in supertree construction and such clades should be collapsed as they have no support [ 41 ]. To demonstrate where the MRP method has performed badly we have included the novel clades in all figures and list details in the figure legends. In addition, 58 nodes are supported by only one character (see additional file 1 ). Each of these nodes is left over from a single source tree. Assessing the support for such nodes is problematic because this may simply reflect a lack of research directed at the taxa in question. A major challenge for supertree construction is to develop measures of support that reflect the robustness of nodes in the source trees. We list the number of characters supporting each node ( additional file 1 ) but stress that these are not measures of tree robustness and may not be directly comparable even within the same tree. This is because the taxon coverage across source trees is highly variable so some nodes have more potential support than others. Furthermore, because measures of support used in the source trees differ between studies (some source trees include no measures of support), it is impractical and of dubious value to use these measures to assess the robustness of the supertree. Figure 2 Summary of shorebird supertree. Family and subfamily level relationships of shorebirds based on 50% majority rule tree. Numbers on nodes refer to age estimates in additional file 1. Boxed node numbers indicate that node collapses to its immediate ancestor in the strict consensus tree (see also additional files 2 and 3 for the full 50% majority rule and strict consensus trees respectively). Figure 3 Phylogeny of Larini. 50% majority rule supertree showing the relationships of the Larini. Numbers on nodes refer to age estimates in additional file 1. Boxed node numbers indicate that node collapses to its immediate ancestor in the strict consensus tree (see also additional files 2 and 3 for the full 50% majority rule and strict consensus trees respectively). Figure 4 Phylogeny of Sternini. 50% majority rule supertree showing the relationships of the Sternini. Numbers on nodes refer to age estimates in additional file 1. Boxed node numbers indicate that node collapses to its immediate ancestor in the strict consensus tree (see also additional files 2 and 3 for the full 50% majority rule and strict consensus trees respectively). Figure 5 Phylogeny of Rynchopini, Stercorariini, Dromas , Alcinae, and Glareolidae 50% majority rule supertree showing the relationships of the Rynchopini, Stercorariini, Dromas , Alcinae, and Glareolidae. Numbers on nodes refer to age estimates in additional file 1. Boxed node numbers indicate that node collapses to its immediate ancestor in the strict consensus tree (see also additional files 2 and 3 for the full 50% majority rule and strict consensus trees respectively). Node numbers 139 and 140 have no support from any source tree and are novel clades. Figure 6 Phylogeny of Jacanidae, Rostratulidae, Thinocoridae, Pedionomidae and Scolopacidae 50% majority rule supertree showing the relationships of the Jacanidae, Rostratulidae, Thinocoridae, Pedionomidae and Scolopacidae. Numbers on nodes refer to age estimates in additional file 1. Boxed node numbers indicate that node collapses to its immediate ancestor in the strict consensus tree (see also additional files 2 and 3 for the full 50% majority rule and strict consensus trees respectively). Node numbers 85 and 89 have no support from any source tree and are novel clades. Figure 7 Phylogeny of Scolopacidae 50% majority rule supertree showing the relationships of the Scolopacidae. Numbers on nodes refer to age estimates in additional file 1. Boxed node numbers indicate that node collapses to its immediate ancestor in the strict consensus tree (see also additional files 2 and 3 for the full 50% majority rule and strict consensus trees respectively). Node numbers 108 and 122 have no support from any source tree and are novel clades. Figure 8 Phylogeny of Pluvianellidae, Chionidae, Burhinidae, Haematopodini and Recurvirostrini 50% majority rule supertree showing the relationships of the Pluvianellidae, Chionidae, Burhinidae, Haematopodini and Recurvirostrini. Numbers on nodes refer to age estimates in additional file 1. Boxed node numbers indicate that node collapses to its immediate ancestor in the strict consensus tree (see also additional files 2 and 3 for the full 50% majority rule and strict consensus trees respectively). Node numbers 20 and 29 have no support from any source tree and are novel clades. Figure 9 Phylogeny Charadriinae 50% majority rule supertree showing the relationships of the Charadriinae. Numbers on nodes refer to age estimates in additional file 1. Boxed node numbers indicate that node collapses to its immediate ancestor in the strict consensus tree (see also additional files 2 and 3 for the full 50% majority rule and strict consensus trees respectively). Node number 57 have no support from any source tree and are novel clades. The majority of unresolved nodes in the shorebird supertree are located towards the tips of the phylogeny. For example, the genus Gallinago forms a monophyletic clade but only two pairs of species are resolved from 14 species ( G. megala and G. negripennis ; G. macrodactyla and G. media ) in the majority-rule tree. Only the latter relationship remains in the strict consensus tree. In addition, clades including the genera Charadrius and Vanellus , Calidris and Tringa , Sterna , and Scolopax are poorly resolved. This may reflect a bias in phylogenetic studies of shorebirds. For instance, we found six source trees for Alcinae [ 43 - 48 ] but none devoted to Scolopax or Gallinago . Thomas et al. [ 49 ] indicate that this may be a problem for shorebird studies in general and reported a strong skew favouring research on northern hemisphere species. In contrast to the within genera relationships, the generic and family levels are generally well resolved. The supertree indicates three monophyletic Charadriiformes lineages (figure 2 ). Family and subfamily resolution within each lineage is high, however the relative position of each group is unresolved in the strict consensus tree. This is an important point because the deepest relationships of shorebird phylogeny are contentious [ 22 ]. The 50% majority-rule consensus tree indicates that the gulls and allies (Larini, Sternini, Rynchopini, Stercorariini, Dromas , Alcinae, and Glareolidae) are sister to the sandpipers and allies (Scolopacidae, Jacanidae, Rostratulidae, Thinocoridae, Pedionomidae). The most basal lineage includes the plovers and allies (Charadriinae, Pluvianellidae, Chionidae, Burhinidae, Haematopodini and Recurvirostrini). The gulls and allies clade is most consistent with DNA-DNA hybridisation [ 16 ], indicating that Larini are sister to Sternini and that Rynchopini are sister to this group. This conflicts with morphology-based topologies where Stercorariini are sister to Larini and Sternini with Rynchopini basal to both. Indeed, the position of Stercorariini remains controversial and most recently they were placed as sister to Alcinae [ 20 - 22 ]. In contrast, morphological evidence [ 18 , 19 ] places Alcinae at the base of the whole Charadriiformes tree with Stercorariini sister to Larini. Thus, the position of Alcinae is uncertain and appears to be dependent on the type of data, with fundamental differences between molecular based analyses and morphological analyses. The taxon sampling of previous morphological and molecular studies varies considerably and it may be this, rather than genuine differences in the phylogenetic signal of different data types, that is the cause of conflict in resolving the phylogenetic position of Alcinae. However, it is encouraging that van Tuinen et al. [ 17 ] suggested that new unpublished osteological data are consistent with the more derived position indicated by molecular data. The supertree resolves Glareolidae outside the Larini, Sternini, Rynchopini, Stercorariini, Dromas , Alcinae clade. This is also the case with recent molecular and previous DNA-DNA hybridisation studies. Morphological studies have failed to resolve the position of Glareolidae, placing the family in a large polytomy with all other major groups except Alcinae and the sandpipers and allies (fig. 1 ). A novel development in shorebird phylogeny is the placement of the black-rumped buttonquail Turnix hottentotta as a sister to the gulls and allies (Larini, Sternini, Rynchopini, Stercorariini, Dromas , Alcinae, and Glareolidae) based on the nuclear RAG-1 gene [ 20 ]. We did not include this species in the supertree because to date Paton et al. [ 20 ] remains the only study to reveal an apparently robust relationship. More diverse sampling of the buttonquails (Turnicidae) is essential to corroborate the general affinities of this family. The relationships within the plover clade appear to be reasonably stable. Morphological, molecular, and DNA-DNA hybridisation all place Charadriinae as sister to Haematopodini and Recurvirostrini; our supertree is consistent with these relationships. However, it is not clear whether Burhinidae and Chionidae are sister to each other [ 20 - 22 ] or whether Chionidae are sister to a Charadriinae, Haematopodini, Recurvirostrini, and Burhinidae clade [ 16 ]. Our supertree also included Pluvianellidae, a family consisting of only one species (magellanic plover Pluvianellus socialis ) and places this as sister to Chionidae. If Pluvianellidae are excluded, the supertree is consistent with the sister group relationship of Burhinidae and Chionidae. The sister group relationship of Jacanidae to Rostratulidae is well established [ 16 , 18 - 22 ] and is found in our supertree. The supertree resolves the Thinocoridae and Pedionomidae as sister taxa and this group is sister to the Jacanidae and Rostratulidae. The large Scolopacidae clade is at the base of the sandpiper clade consistent with recent molecular studies [ 20 - 22 ] and the DNA-DNA hybridisation tapestry [ 16 ]. Taken together, it is evident that the supertree is generally more consistent with molecular data (both recent sequence studies and DNA-DNA hybridisation) than with analyses based on morphology. However, it is of course possible that this reflects the greater number of molecular source trees available rather than indicating that molecular data is actually better at resolving shorebird phylogeny. We included several large morphological phylogenies [e.g [ 18 , 19 , 26 , 30 , 43 ]] but the majority of source trees (29 out of 51) were based on molecular evidence (see additional file 5 ). Node dates The higher resolution of the majority-rule tree means it is more likely to be of use in comparative studies. We therefore estimated node ages for this topology only (see additional file 1 and 2 ). We stress that our estimates of node dates are a first attempt at dating the whole tree and have several limitations. First, the fossils used to calibrate seven nodes in the tree are unlikely to be the earliest members of their respective families thus these dates will be underestimates. Second, we assumed that the fossils are grouped with the extant members of the family but this requires formal testing in a phylogenetic framework. Third, the pure birth model assumes that no extinction occurs but this may be unrealistic and it is likely that extinction processes have reduced the representation of older lineages [ 15 ]. Furthermore, this model is derived from the topological structure of the tree so errors in tree reconstruction will likely lead to errors in branch length estimation. However, this approach has been employed previously in supertrees of primates [ 39 ] and carnivores [ 15 ] explicitly to facilitate comparative analyses. Despite these caveats, simulation studies have demonstrated that comparative methods such as independent contrasts are robust to errors in branch length [ 50 ] and no viable alternative for dating supertrees has been proposed. Nonetheless, we urge that alternative branch length assumptions are explored if the shorebird supertree is used in future comparative studies. At present, the calibrated RAG-1 tree of Paton et al. [ 20 ] remains arguably the most thorough and reliable measure of divergence times for Charadriiformes. A fuller understanding of the phylogenetic affinities of fossil shorebirds will probably improve estimates of node ages for the group. For example, the extinct form Graculavidae, is represented by fossils from the Maastrichtian of New Jersey [ 51 ] and Cretaceous of Wyoming [ 52 ] but its position within the shorebird clade is unclear. Feduccia [ 53 ] suggests that it may be basal and a formal corroboration of this would support proposals for a late Cretaceous origin of shorebirds. The difficulties in dating the shorebird tree are further illustrated by fossil representatives of Recurvirostrini and Burhinidae which are much older than current estimates suggests. The earliest record of the Recurvirostrini is estimated to be over 50 million years old [ 54 ] whilst recent discoveries of a possible member of the Burhinidae are dated to around 70 mya [ 55 , 56 ]. There is clearly a need for an integrated phylogenetic study including both extinct and extant shorebirds. Supertree bias Supertrees are still at an early stage of development and many aspects of MRP, and supertree methods in general, are not yet clearly understood. Steps can be taken to ensure that the supertree includes the most appropriate sets of sources trees, such as only using trees from explicitly phylogenetic studies. This is not always straightforward and could result in the exclusion of important information. For instance, in our shorebird supertree, we included Sibley and Ahlquist's DNA-DNA hybridisation tapestry [ 16 ] although this is based on distance measures rather than more rigorous phylogenetic methods. Even if very strict tree selection criteria are applied, there are still likely to be biases in the data set. For example, not all source trees are equally well supported, yet in most supertree analyses each tree is treated equally [ 57 ]. This is a problem for supertree construction because whilst it is theoretically possible, and indeed beneficial, to weight source trees based on support values [ 57 ] it is rarely possible in practice. Many source trees do not have support values and those that do may use different methods, (e.g, bootstrapping or decay indices) which cannot be directly compared with each other. An additional problem that has not been fully resolved relates to correlations between source trees [ 58 ]. Several source trees based on the same data set may unduly increase the influence of that data set on the supertree analysis. However, there is no formal way of determining how much overlap to allow and the choice of source trees that go into supertree construction inevitably involves some degree of subjective reasoning. For the shorebird supertree we used strict Reduced Cladistic Consensus trees to summarise potential source trees that were from the same data set but based on different methods. For example, Thomas et al. [ 22 ] based their phylogeny on cytochrome- b but used a range of methods including parsimony and Bayesian analyses. We therefore combined these trees to minimise bias. In contrast, Ericson et al. [ 21 ] used two types of data: sequences from the nuclear RAG 1 gene and sequences from the myoglobin intron II. They carried out three analyses: each gene separately and then the two combined in a single analysis. In this case, we used three source trees. It could be argued that the combined analysis of Ericson et al. [ 21 ] should be excluded because of the possible overlap with the individual analyses. However, under the principle of total evidence, the combined data set may result in novel relationships being revealed [ 31 , 33 ] and therefore could contribute important information to the supertree. Simulation and empirical studies are required to fully understand these and other possible biases in supertree construction (e.g., the influence of source tree size and shape) and formal protocols for the selection of source trees are desirable. For transparency, we include a summary of the source trees used, data type, and the main taxa included in the study ( additional file 5 ). Our shorebird supertree is highly consistent with recent advances in the molecular phylogenetics Charadriiformes. However, we urge caution when using the tree in comparative analyses and encourage the additional use of alternative phylogenies and branch length assumptions. It is particularly important to note that the position of some groups such as the Alcinae remains controversial and that although the majority rule tree is consistent with recent molecular studies, the strict consensus tree fails to resolve the deepest nodes. Conclusions The supertree presented here is, to our knowledge, the first attempt to reconstruct the phylogeny of the entire order Charadriiformes. Overall, the supertree is highly consistent with recent molecular hypotheses of shorebird phylogeny. However, it is apparent that fresh attempts to resolve both the phylogeny and estimates of age will be dependent on further gene sequencing and new fossil discoveries. The affinities of the Alcinae and the relationships between the three major shorebird clades require further corroboration, and studies of several genera such as Gallinago and Vanellus are desirable. Furthermore, additional work is required to establish the true affinities of the Turnicidae. Nonetheless, it appears that shorebird phylogeny is gradually approaching a consensus view. The broad taxonomic scope and consistency of the supertree mean that is of potentially great value to future comparative studies (accepting the caveats discussed above) of the behaviour, life-history, ecology and conservation of this diverse group. Methods Supertree construction Possible source trees were identified from online searches of Web of Science covering the years 1981 to 2004. We used the single key strings phylogen, cladistic, clado, classif, systematic, and taxonom (where the asterisks allow variations such as "phylogeny" or "phylogenetics") in the topic field, in conjunction with a major Charadriiformes taxon name (scientific or common). As supertree methods have been criticized for being biased towards historical trends, we preferred those studies that explicitly set out to derive a phylogenetic hypothesis and so exclude purely (and typically older) descriptive taxonomic works. The Sibley and Ahlquist [ 16 ] DNA-DNA hybridisation tapestry may be viewed as non-cladistic, but it was clearly the authors' intention to reconstruct the phylogeny of birds. Furthermore, it provided a vital catalyst for subsequent studies of avian (including shorebird) phylogeny. We therefore included the DNA-DNA hybridisation hypothesis as a source tree in our analyses. Simulation studies have demonstrated that the performance of supertree methods is improved by including at least one taxonomically complete (or near complete) source tree [ 57 ]. We therefore make an exception to our self-imposed rule, and in addition use the taxonomic hierarchy of Monroe and Sibley [ 1 ] as a source tree as this includes all extant Charadriiformes species. We acknowledge that this taxonomy is based largely on Sibley and Ahlquist's [ 16 ] DNA-DNA hybridisation tapestry. The initial search identified 78 source trees from 44 publications. Each source tree was typed as a text file in Nexus format [ 59 ]. We coded trees to the species level with species names taken from Monroe and Sibley [ 1 ], but note that contra Monroe and Sibley [ 1 ], we use Charadriiformes not Charadrii to refer to the whole group. Several studies included the gull Larus thayeri [ 26 , 60 - 63 ] either as a subspecies of Larus glaucoides ( Larus glaucoides thayeri in Monroe and Sibley [ 1 ]) or a species in its own right. In recognition of this, we included Larus glaucoides thayeri as the only subspecies in our data set thus increasing the total taxa to 366. Monroe and Sibley [ 1 ] include 16 species of the family Pteroclidae within the Charadriiformes. However, the relationship of this family to the Charadriiformes is uncertain and they have recently been placed in their own order [ 64 ]. We include the Pteroclidae in our analyses only as a means of rooting the tree. Where there were multiple most parsimonious trees (MPTs), or where source trees had been derived from predominantly overlapping data (e.g., from the same data but using alternative methods), we used RadCon [ 65 ] to produce strict Reduced Cladistic Consensus trees (RCC [ 66 , 67 ]). The output is in the form of a reduced consensus profile and from this we selected the tree with the highest Cladistic Information Content (CIC) [ 65 , 68 ]. This resulted in a total of 51 source trees from which our supertree is derived and these are summarised in additional file 5 . We produced an MRP matrix of the 51 Nexus [ 59 ] source trees in RadCon [ 65 ] (see additional file 6 for the MRP file). We used the original MRP coding method of Baum [ 37 ] and Ragan [ 38 ]. Weighting source trees based on node support such as bootstrapping improves the accuracy of MRP supertrees [ 57 ]. However, this is only possible if all source trees can be weighted on the same criteria [ 57 ]. The absence of branch support measures in many of the shorebird source trees precludes this approach from the present study; hence, subsequent analyses were conducted using equally weighted parsimony. The tendency of large data sets to produce many sub-optimal trees that are close in length and topology to the shortest tree is a serious problem in phylogenetics. Standard heuristic searches frequently are trapped searching within globally sub-optimal "islands" and the tree search is often aborted before completion. Nixon [ 69 ] proposed a new method to avoid this problem. The "Parsimony Ratchet" reweights a random set of characters from the data set. This may result in the tree island no longer representing a local optimum and the heuristic search continues until a new optimum is reached. The algorithm then reverts to the original weighting and the search continues. Nixon [ 69 ] demonstrated the efficacy of the method on a 500-taxon data set, where the ratchet-based search found a tree two steps shorter than standard heuristic searches. We used PAUPRat [ 70 ] to implement a parsimony ratchet in PAUP* [ 59 ]. The default settings of 200 iterations and 15% perturbation of characters for reweighting were used and we carried out 20 replicates. Equally parsimonious trees were summarized using both strict and 50% majority-rule consensus methods. We did not calculate any measures of branch support for two reasons. First, their validity and meaning is questionable in MRP supertrees [ 41 ]. Second, the number of taxa included in our data set is too large to allow practical calculation of any branch support indices (e.g., decay indices [ 71 ]) on a desktop computer. Dating the supertree Following Purvis [ 39 ] and Bininda-Emonds et al. [ 15 ] we dated the supertree using both absolute and relative dates. We used data from the Fossil Record 2 [ 54 ] as the source of fossil-based absolute dates. This yielded estimates for Jacanidae ( Nupharanassa tolutaria , Rupellian), Phalaropus ( Phalaropus elenorae , Middle Pliocene), Burhinidae ( Burhinus lucorum , Lower Miocene), Glareolidae ( Paractiornis perpusillus , Lower Miocene), Alcinae ( Petralca austrica , Rupellian), Stercoariini ( Stercorarius sp., Middle Miocene), and Larini (undetermined, Rupellian). We took the midpoint of the range from the Fossil Record 2 [ 54 ] as our date estimate. More recent publications of fossil Charadriiformes were not included because they either represent specimens that are younger or have not been assigned to families that are represented amongst the extant Charadriiformes (such as Turnipacidae [ 72 ]). We assumed that fossil dates represent the earliest occurrence for each group which inevitably resulted in underestimates of clade age. The fossil record of Charadriiformes is amongst the best of the modern bird groups [ 17 ] in terms of the numbers of taxa, but many specimens are fragmentary and reliable estimates of divergence dates are dependent on a limited number of exceptional specimens [ 73 ]. The phylogenetic affinities of the fossil shorebirds in relation to their extant relatives have not yet been fully established, hence have implicitly assumed that fossil representatives of extant groups would be resolved amongst their living relatives. Source trees may include estimates of relative branch lengths (e.g., genetic distances). This allows further dating of the supertree but is problematic because different relative estimates are not comparable and cannot be applied directly to the supertree [ 39 ]. However, where a source trees shares a node that has an absolute date in the supertree (a node dated from fossil evidence), the relative branch lengths can easily be converted to estimates of age. All taxa in our supertree are either extant, or very recently extinct; hence, the tips of the calibrated supertree should be equidistant from the root of the tree. In source trees where the relative branch lengths are not equidistant from the root, we followed the protocol of Purvis [[ 39 ]; p.407–8]. We estimated relative dates using the local molecular clock logic [ 74 ] as implemented by Purvis [ 39 ] and Bininda-Emonds et al. [ 15 ]. For example, consider three taxa A, B, and C where A and B are sister taxa and C is sister to A and B. The root is dated to 10 million years (myr) from fossil evidence, and independent molecular data provides estimates of divergence based on the number of substitutions per site. The molecular estimates of branch lengths are as follows: A , 6 substitutions; B , 8 substitutions; C , 20 substitutions; A and B are 11 substitutions from the root. A and B are therefore separated from their common node by a mean of 7 substitutions. The total length from A and B to the root is thus 18 substitutions compared to 20 for C (a mean of 19). This can be converted to date estimates such that 19 substitutions are equivalent to 10 myr. The dates of the tree are then: (( A : 3.68, B : 3.68), C : 10)). There were no cases where multiple source trees with molecular divergence dates were able to provide estimates for the same node. We estimated relative dates from multiple nodes rather than a single dated node to minimise correlative errors in estimates. To provide date estimates for all nodes in the tree we employed a pure birth model to date nodes for which absolute and relative dates could not be attained [ 39 ]. Pure birth models infer that a clade's age is proportional to the logarithm of the number of species within the clade: date of daughter = date of ancestor (log daughter clade size/log parent clade size) For example, the age of a daughter node that subtends 12 taxa, estimated from its immediate ancestor dated to 20 myr and which subtends 19 taxa is: 20(log(12)/log(19)) = 16.879 We applied this approach to estimate the ages of daughter nodes based on dates (absolute or calibrated) of ancestral nodes. We had no ancestral node on which to base estimates of the most basal clade. In this case, we rearranged the pure birth formula and calculated the age of the ancestral node from its two daughter nodes, taking the mean as our "best estimate". Finally, to estimate the ages of nodes between daughter and ancestor nodes of known age we spaced the nodes evenly along the branches length [ 75 ]. Authors' contributions GHT assisted in the design of the study, carried out the phylogenetic analyses and node dating, and drafted the manuscript in partial fulfillment of a doctoral degree at the University of Bath. MAW assisted in the design of the study and with editing and revision of the manuscript. TS assisted in the design of the study, collection of source trees, and editing and revision of the manuscript. All authors read and approved the final manuscript. Supplementary Material Additional File 1 Estimates of node ages and node support (branch lengths.xls) Node numbers correspond to figures 2-9. Five types of estimate were used: a) absolute dates from the fossil record; b) absolute dates from molecular point estimates; c) relative dates based on branch length estimates from molecular studies; d) estimates based on a pure birth model (see text for details); and e) even spacing of nodes along branches with daughters and ancestors of known age. The numbers of characters supporting each node are provided (column D), this is equivalent to the number of source trees that share the equivalent node (see text for details). Click here for file Additional File 2 Shorebird supertree (50% majority-rule consensus; majrulesupertree.tiff) Shorebird supertree based on 50% majority-rule consensus of 1496 shortest trees with calibrated branch lengths. Scale bar indicates time from the present in millions of years. Click here for file Additional File 3 Shorebird supertree (strict consensus; strictsupertree.tif) Shorebird supertree based on 50% majority-rule consensus of 1496 shortest trees. Click here for file Additional File 5 Source trees (source trees.xls) A summary of each tree used is given including the data type and main taxa studied. This is a brief summary and the original papers should be consulted for full details. Click here for file Additional File 6 MRP matrix (shorebirdMRP.txt) The MRP matrix used in the shorebird supertree analysis. Click here for file Additional File 4 Calibrated supertree (shorebirdsupertree.txt) The supertree in nexus format including branch length estimates. Click here for file	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC515296.xml
528856	Funding free and universal access to Journal of Neuroinflammation	Journal of Neuroinflammation is an Open Access, online journal published by BioMed Central. Open Access publishing provides instant and universal availability of published work to any potential reader, worldwide, completely free of subscriptions, passwords, and charges. Further, authors retain copyright for their work, facilitating its dissemination. Open Access publishing is made possible by article-processing charges assessed "on the front end" to authors, their institutions, or their funding agencies. Beginning November 1, 2004, the Journal of Neuroinflammation will introduce article-processing charges of around US$525 for accepted articles. This charge will be waived for authors from institutions that are BioMed Central members, and in additional cases for reasons of genuine financial hardship. These article-processing charges pay for an electronic submission process that facilitates efficient and thorough peer review, for publication costs involved in providing the article freely and universally accessible in various formats online, and for the processes required for the article's inclusion in PubMed and its archiving in PubMed Central, e-Depot, Potsdam and INIST. There is no remuneration of any kind provided to the Editors-in-Chief, to any members of the Editorial Board, or to peer reviewers; all of whose work is entirely voluntary. Our article-processing charge is less than charges frequently levied by traditional journals: the Journal of Neuroinflammation does not levy any additional page or color charges on top of this fee, and there are no reprint costs as publication-quality pdf files are provided, free, for distribution in lieu of reprints. Our article-processing charge will enable full, immediate, and continued Open Access for all work published in Journal of Neuroinflammation . The benefits from such Open Access will accrue to readers, through unrestricted access; to authors, through the widest possible dissemination of their work; and to science and society in general, through facilitation of information availability and scientific advancement.	Introduction Journal of Neuroinflammation is an Open Access, online journal that is published by BioMed Central, an independent publisher committed to Open Access for peer-reviewed biomedical research [ 1 ]. Among the many benefits of Open Access publishing are (1) instant and universal availability of published work to any potential reader, worldwide, completely free of subscriptions, passwords, and charges; and (2) copyright retention by the authors rather than the publisher. This and many other benefits led us to select Open Access publishing, and to select BioMed Central, for the Journal of Neuroinflammation . Open Access publishing is made possible by article-processing charges (APCs) assessed "on the front end" to authors, their institutions, or their funding agencies. The Journal of Neuroinflammation will introduce APCs of around US$525 per article for manuscripts submitted on or after November 1, 2004. This charge will be waived for all authors from institutions that are BioMed Central members, and in additional cases for reasons of genuine financial hardship. Problems with the traditional publishing model Traditional journals generally do not charge authors for publication (although assessed page or color charges may easily exceed our APCs). Instead, article access is traditionally paid for by readers, either through subscriptions or through fees assessed for online viewing and downloading. Over the past decade, escalating journal subscriptions have resulted in cash-strapped libraries cancelling journal subscriptions [ 2 ], thus limiting the range of articles available to many readers and limiting the potential audience available to authors. The Open Access publishing model The Journal of Neuroinflammation's Open Access policy changes the way in which articles are published. First, all articles become freely and universally accessible online, immediately upon acceptance, so an author's work can be read by anyone at no cost. Second, the article authors retain copyright for their work, and grant to anyone the right to reproduce and disseminate the article, provided that it is correctly cited and no errors are introduced [ 1 ]. Third, a copy of the full text of each article is permanently archived in several separate online repositories. Journal of Neuroinflammation's articles are permanently archived in PubMed Central [ 3 ], the US National Library of Medicine's full-text repository of life science literature, and also in repositories at the University of Potsdam [ 4 ] in Germany, at INIST [ 5 ] in France and in e-Depot [ 6 ], the National Library of the Netherlands' digital archive of all electronic publications. Benefits of Open Access publishing Open Access has four broad benefits for science and the general public. First, published work is disseminated freely and instantly to the widest possible audience, without barriers to access. Authors are free to reproduce and distribute their work at will, for example by placing it on their institution's website. Open Access publication has been shown to actually increase article citations and impact because of this easier availability [ 7 ]. Second, availability of Open Access articles enhances literature searching [ 8 ], as information available to researchers is not limited by what their libraries can afford. Third, results of publicly funded research are accessible to all taxpayers and not just those with access to libraries with journal subscriptions. Such public accessibility would actually become a legal requirement in the USA if the proposed Public Access to Science Act is enacted into law [ 9 ]. Fourth, article access is not limited by the economic resources of a scientist's country or institution; resource-poor countries and institutions are able to access the same material as wealthier ones, subject only to the availability of internet access [ 10 ]. Journal of Neuroinflammation's article-processing charges Article-processing charges will allow continued Open Access to all article published in Journal of Neuroinflammation . Authors will be asked to pay around US$525 upon acceptance of their article for publication. Submitted articles that are not accepted will incur no charge. There will be no charges for authors from institutions that are institutional members of BioMed Central. Currently this includes NHS England and all universities in the UK, the US National Institutes of Health and 136 other institutions and universities in the USA, the World Health Organization, and almost 200 additional institutions in 37 other countries [ 11 ]. Potential authors who are not associated with these institutions can avoid article-processing charges by getting their institution to join this list of BioMed Central institutional members. The annual institutional membership fee covers APCs for all authors at that institution for that year. In addition, many funding agencies have recognized the importance of Open Access publishing and have specified that funds from their grants may be used directly to pay APCs [ 12 ]. Finally, APC waivers are available for cases of genuine financial hardship. These will be considered on a case-by-case basis by the Editors-in-Chief. What do article-processing charges pay for? The APC pays for an electronic submission process that facilitates efficient and thorough peer review, for publication costs involved in providing the article freely and universally accessible in various formats online, and for the processes required for the article's inclusion in PubMed and its archiving in PubMed Central, e-Depot, Potsdam and INIST. There is no remuneration of any kind provided to the Editors-in-Chief, to any members of the editorial board, or to peer reviewers; all of whose work is entirely voluntary. Although some authors may consider US$525 expensive, it must be remembered that Journal of Neuroinflammation does not levy any additional page or color charges on top of this fee. Because we are an online-only journal, any number of color figures, photographs, and 'extra' pages can be included at no extra cost. Such color and page charges, as assessed by more traditional journals, can easily exceed our flat US$525 per-article APC. Another common expense with traditional journals is the purchase of reprints for distribution, and the cost of these reprints is also frequently greater than our APCs. The Journal of Neuroinflammation provides free, publication-quality pdf files for distribution, in lieu of reprints. Free access versus Open Access Several traditional journals now offer free access to their articles online, but this is different from Open Access as defined by the Bethesda Statement [ 13 ]. First, this access may be delayed for 6–2 months after publication. Second, readers are not free to reproduce and/or disseminate the work because of restrictions imposed by publishers' copyright policies. Even these restrictive policies do not ensure continued free access; the British Medical Journal , for instance, recently announced that it cannot continue to provide free access to its website [ 14 ]. They are considering various sources of revenue, including APCs [ 15 ]. APC-funded Open Access is not unique to BioMed Central or to the Journal of Neuroinflammation . The USA-based Public Library of Science (PLoS) is a new, non-profit organization that, like BioMed Central, is dedicated to online, Open Access publishing. PLoS has started two new Open Access journals, with APCs of US$1500 for each accepted article [ 16 ]. PLoS has used television advertising to promote their new journals [ 9 ], providing a high profile that should raise awareness of Open Access publishing in general. This, in turn, should encourage researchers in all disciplines to understand and accept Open Access, and to accept APCs as an acceptable funding method. Conclusion Article-processing charges will enable full, immediate, and continued Open Access for all work published in Journal of Neuroinflammation . The benefits from such Open Access will accrue to readers, through unrestricted access; to authors, through the widest possible dissemination of their work; and to science and society in general, through facilitation of information availability and scientific advancement. We ask for your support in this important movement by submitting your next article to Journal of Neuroinflammation or to another Open Access journal. Competing interests At Journal of Neuroinflammation, the work of the Editors-in-Chief, the Editorial Board, and of all invited outside peer reviewers is entirely voluntary, without tangible remuneration of any kind. Our goal is publication of biomedical research of the highest quality, and our (intangible) rewards lie in the achievement of these goals. Decisions about manuscripts are based entirely on the quality of the work, and not on the ability of authors to pay article-processing charges. Abbreviations APC = article-processing charge.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC528856.xml
521682	Differential dynamics of histone H3 methylation at positions K4 and K9 in the mouse zygote	Background In the mouse zygote the paternal genome undergoes dramatic structural and epigenetic changes. Chromosomes are decondensed, protamines replaced by histones and DNA is rapidly and actively demethylated. The epigenetic asymmetry between parental genomes remains at least until the 2-cell stage suggesting functional differences between paternal and maternal genomes during early cleavage stages. Results Here we analyzed the timing of histone deposition on the paternal pronucleus and the dynamics of histone H3 methylation (H3/K4mono-, H3/K4tri- and H3/K9di-methylation) immediately after fertilization. Whereas maternal chromatin maintains all types of histone H3 methylation throughout the zygotic development, paternal chromosomes acquire new and unmodified histones shortly after fertilization. In the following hours we observe a gradual increase in H3/K4mono-methylation whereas H3/K4tri-methylation is not present before latest pronuclear stages. Histone H3/K9di-methylation is completely absent from the paternal pronucleus, including metaphase chromosomes of the first mitotic stage. Conclusion Parallel to the epigenetic asymmetry in DNA methylation, chromatin modifications are also different between both parental genomes in the very first hours post fertilization. Whereas methylation at H3/K4 gradually becomes similar between both genomes, H3/K9 methylation remains asymmetric.	Background It is now generally accepted that the properties of a particular DNA sequence in cells are not solely defined by the nucleotide sequence itself, but by "epigenetic" modifications as well. Epigenetic modifications imply the methylation of cytosine residues in CpG dinucleotides and covalent modifications of core histones. These modifications allow for flexible, but heritable at the same time, reprogramming of the genome. In histone H3 five lysine residues can be methylated (K4, K9, K27, K36 and K79) [ 1 ]. Methylation at K4 and K9 play opposite roles in structuring repressive or accessible chromatin domains, with K4 methylation associated with transcriptionally active chromatin and K9 methylation with inactive chromatin in higher eukaryotes [ 2 ]. In addition, these lysine residues can be mono-, di- or tri-methylated, which contributes to the distinct qualities of H3/K4 and H3/K9 methylation. Similar to H3/K9 methylation, DNA methylation is associated with silenced chromatin and there appeared to be an interplay between the two epigenetic modifications. It is still an open question whether DNA methylation directs H3/K9 methylation or other way around, for both scenarios the experimental evidences do exist [ 3 , 4 ]. It recently has been shown that in mammalian cells the maintenance DNA methyltransferase DNMT1 is associated with proteins involved in chromatin reprogramming, including histones deacetylases, and is required for the establishment of H3/K9 methylation [ 5 ]. Various experimental data suggest that the DNA methylation causes multiple changes in local nucleosomes, such as deacetylation of histones H3 and H4, prevents H3/K4 methylation and induces H3/K9 methylation [ 6 ]. The fertilization of mouse egg causes dramatic changes in organization of both paternal and maternal genomes. Initially arrested in metaphase II oocyte completes the meiosis, forming haploid maternal pronucleus and extruding the second polar body. The densely packed with protamines sperm DNA decondences, protamines get exchanged by histones and DNA undergoes active demethylation. The demethylation in the early mouse zygote occurs asymmetrically on paternal DNA and affects different classes of repetitive and single copy sequences, but not the control regions of imprinted genes [ 7 , 8 ]. Previous studies have shown the exclusive localization of methylated H3/K9 in maternal pronucleus of the mouse zygote, which additionally marks the epigenetic asymmetry between maternal and paternal pronuclei [ 9 - 11 ]. Here we examine time dependent changes of chromatin structure in the mouse zygote, focusing on the dynamics of the acquisition of histones in the paternal pronucleus and methylation status of histone H3 at positions K4 and K9. Results and discussion In order to obtain mouse zygotes at different stages of development and to provide the precise timing for fertilization we used in vitro fertilization of mature mouse oocytes. Histones and methylated H3/K4 and H3/K9 were detected by using indirect immunofluorescence. In our experiments we used antibodies, which specifically recognize mono- or tri-methylated forms of H3/K4, and di-methylated H3/K9. The zygotes were analyzed after 3, 5, 8, 10, 12 and 18 hours incubation of mature oocytes with capacitated sperm from donor males. After 18 hours most of embryos were found to be at metaphase and some already at telophase stage of the first mitotic division. Even using in vitro fertilization, the obtained zygotes are not completely synchronous in their development and it is more appropriate to use PN stages classification, which is based on the morphological changes of both pronuclei [ 8 , 12 ]. Appearance of histones on paternal chromosomes We performed the immunostaining against core histones (anti-PanHistone antibodies) in all the stages tested in combination with antibodies, recognizing the specific methylated forms of histone H3. This served as a positive control for the immunostaining procedure and allowed us to follow the dynamics of histone acquisition in the paternal pronucleus. Histones were first detected shortly after the penetration of sperm into the oocyte and the beginning of the decondensation of sperm chromatin. According to PN stages classification we could clearly detect histones on paternal pronucleus at late PN 0 /early PN 1 stages (approx. 3–5 hours p.f.), exactly when the global demethylation starts [ 8 ] (Fig. 1 ). Figure 1 Dynamic changes in chromatin of zygotes at different pronuclear stages. DNA is visualized by DAPI (blue colour) staining. Mouse monoclonalanti PanHistones antibodies were detected by fluorescein conjugated anti-mouse secondary antibodies (green colour). Specific rabbit polyclonal antibodies, recognizing H3/K4monoMe (a), H3/K4triMe (b) or H3/K9diMe (c) were detected by Rhodamine Red-X conjugated anti-rabbit secondary antibodies (red colour). Dynamic changes in H3/K4 methylation in paternal genome Probing the mouse zygotes at different stages with antibodies specifically recognizing either mono- or tri-methylated H3/K4 revealed that these types of modifications are associated with maternal genome through all zygotic stages, including mature oocyte and seem to be rather ubiquitous (Fig. 1a,1b ). As for the paternal pronucleus – we detect the appearance of H3/K4mono-methylation in the beginning of PN 1 (approx. 5 hours p.f.) stage (Fig. 1a ), only slightly delayed compared to the appearance of core histones (Fig. 2 ). By PN 3 – PN 4 stages both paternal and maternal pronuclei show equal staining intensity. This indicates that H3/K4 specific histone methyltransferase, possibly Set9 [ 13 ], is quite active in the early zygote and methylates histone H3 after it is incorporated into the nucleosomes. In contrast to that, it has been shown recently that H3/K9 specific histone methyltransferase is inactivated immediately after the fertilization by yet unknown active mechanism, which involves de novo synthesis of some specific factors [ 11 ]. H3/K4tri-methylation becomes detectable later, starting from PN 4 stage (approx. 8–10 hours p.f.) and the difference in antibodies staining intensity between paternal and maternal pronuclei becomes indistinguishable in the last pronuclear stage PN 5 (approx 12 hours p.f.) (Fig. 1b ) and in metaphase stage of first mitosis approximately 16 hours p.f. (Fig. 3a ). The fact that H3/K4 first becomes mono-methylated and several hours later tri-methylated suggests progressive methylation of histone H3 at lysine 4. We also suggest that histone H3 gets incorporated into the nucleosomes being unmethylated and then undergoes methylation because we observe first the appearance of histones and then H3/K4mono-methylation. In contrast to that – acetylation of histones H3 and H4 happens before they are incorporated into the nucleosomes, and after the nucleosome assembly they can get deacetylated by histone deacetylases (HDACs) whenever required [ 14 ]. But no histone demethylase has been found so far. Figure 2 Distribution of histones and H3/K4monoMe in the zygotes at late PN 0 stage. At this stage histones (green signal) are detectable in both male (♂) and female (♀) pronuclei, whereas H3/K4monoMe (red signal) is only detectable in female pronucleus and polar body (pb). Figure 3 Distribution of H3/K4triMe and H3/K9diMe in metaphase chromosomes during the latter portion of the first cell cycle. (a) Distribution of H3/K4triMe. Paternally and maternally derived chromosomes show equal staining pattern along the whole length of chromosomes. (b) Distribution of H3/K9diMe. This type of modification is not detectable on paternal chromosomes and in maternal chromosomes is mostly associated with centromeres. H3/K9 methylation but not H3/K4 defines the genomes asymmetry in the mouse zygote In order to compare the patterns of H3/K4 and H3/K9 methylation we performed the immunostaining of mouse zygotes using antibodies, which recognize di-methylated H3/K9. Our results are in the agreement with earlier observations that H3/K9 methylation is only attributed to the maternal genome and is completely absent from the paternal [ 9 - 11 ] (Fig. 1c , Fig. 3b ). In normal somatic cells the absence or disruption of H3/K9 methylation leads to the chromosome instability and affects chromosomes segregation during mitosis [ 15 ]. Therefore the absence of H3/K9 methylation on paternal chromosomes is rather surprising and compromises its role in chromosomes segregation. The epigenetic asymmetry between paternal and maternal genomes is observed till 2-cell stage and is characterized by low levels of DNA methylation and H3/K9 methylation in paternal genome [ 8 , 10 , 11 , 16 ]. In case with H3/K4 methylation – the asymmetry is observed only in the beginning of the zygotic development and is indistinguishable in the metaphase stage of the first mitotic division (Fig. 3a ). Recent data from Liu et al . suggest that H3/K9 methylation does not depend on DNA methylation [ 11 ], but it is only paternal DNA which gets demethylated in the mouse zygote and at the same time it does not have detectable H3/K9 methylation. According to data published by Santos et al . [ 17 , 18 ] DNA demethylation starts at PN 1 stage, i.e. at a time when we first observe the appearance of H3/K4mono-methylation (PN 1 stage, Fig. 1a ), and is completed at PN 3 stage when H3/K4mono-methylation in paternal pronucleus reaches approximately the same level as in the maternal (Fig. 1a ). This fact is raising the question if such a coincidence might indicate that DNA demethylation and the establishment of H3/K4 methylation are interdependent. Demethylation of paternal DNA upon the fertilization is not a universal phenomenon for mammalian species. In bovine zygote paternal DNA becomes only partially demethylated, while in sheep and rabbit zygotes the demethylation is hardly detectable [ 17 , 18 ]. The analysis of chromatin modification in early zygotes of these species might help to get an answer if DNA demethylation depends on, or is directed by the specific chromatin modifications. Conclusions Unlike H3/K9 methylation, methylation of H3/K4 is not attributed only to the maternal genome but appears shortly after the acquisition of histones by paternal pronucleus. The methylation of H3/K4 is progressive and by first mitotic division reaches approximately same level as in maternal genome. Methods In vitro fertilization of mouse oocytes As sperm and oocytes donors we used (C57BL/6 X CBA)F 1 mice. Mature oocytes were collected 14 hours post human chorionic gonadotropin injection according to standard procedures [ 19 ]. Sperm isolation and in vitro fertilization (IVF) procedures were performed as described in [ 20 ]. Briefly: the sperm was isolated from cauda epididimus of donor males and capacitated in pre-gassed HTF medium for 1,5 hours. Isolated oocytes in cumulus cell mass were placed into 100 μl drop of HTF medium with capacitated sperm and incubated in CO 2 incubator for 3, 5, or 8 hours. For longer incubation time the oocytes were incubated with sperm in HTF medium for 8 hours and then transferred into the drop of pre-gasses and pre-warmed M16 medium and incubated further for 2, 4 or 10 hours. Immunofluorescence staining After the removal of zona pellucida by treatment with Acidic Tyrode's solution fertilized oocytes were fixed for 20 min in 3.7% paraformaldehyde in PBS, and permeabilized with 0.2% Triton X-100 in PBS for 10 min at room temperature. The fixed zygotes were blocked overnight at 4°C in 1% BSA, 0.1% Triton X-100 in PBS. After blocking the embryos were incubated in the same solution with either anti PanHistones (mouse polyclonal, Roche), anti mono-methyl H3/K4 (rabbit polyclonal, Abcam), anti tri-methyl H3/K4 (rabbit polyclonal, Abcam) or anti di-methyl H3/K9 (rabbit polyclonal, a gift from T. Jenuwein [21] antibodies at room temperature for 1 hour, followed by several washes and incubation for 1 hour with anti-mouse secondary antibodies coupled with fluorescein (Sigma-Aldrich), and anti-rabbit secondary antibodies coupled with Rhodamine Red-X (Jackson ImmunoResearch Laboratories Inc.). After final washes the zygotes were placed on slides and mounted with a small drop of Vectashield (VectorLab) mounting medium containing 0.5 μg 4,6-diamino-2-phenylindole (DAPI). At least 20 zygotes have been analyzed for each stage of zygotic development. Immunofluorescence microscopy The slides were analyzed on Zeiss Axiovert 200 M inverted microscope equipped with the fluorescence module and B/W digital camera for imaging. The images were captured, pseudocoloured and merged using AxioVision software (Zeiss). Authors' contributions KL conducted the experimental part of the work and co-wrote the manuscript. JW coordinated the study and co-wrote the manuscript. All authors read and approved the final manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC521682.xml
521696	The effect of oxythioquinox exposure on normal human mammary epithelial cell gene expression: A microarray analysis study	Background Inter-individual variation in normal human mammary epithelial cells in response to oxythioquinox (OTQ) is reported. Gene expression signatures resulting from chemical exposures are generally created from analysis of exposures in rat, mouse or other genetically similar animal models, limiting information about inter-individual variations. This study focused on the effect of inter-individual variation in gene expression signatures. Methods Gene expression was studied in primary normal human mammary epithelial cells (NHMECs) derived from four women undergoing reduction mammoplasty [Cooperative Human Tissue Network (National Cancer Institute and National Disease Research Interchange)]. Gene transcription in each cell strain was analyzed using high-density oligonucleotide DNA microarrays (HuGeneFL, Affymetrix™) and changes in the expression of selected genes were verified by real-time polymerase chain reaction at extended time points (ABI). DNA microarrays were hybridized to materials prepared from total RNA that was collected after OTQ treatment for 15, 60 and 120 min. RNA was harvested from the vehicle control (DMSO) at 120 min. The gene expression profile included all genes altered by at least a signal log ratio (SLR) of ± 0.6 and p value ≤ 0.05 in three of four cell strains analyzed. Results RNA species were clustered in various patterns of expression highlighting genes with altered expression in one or more of the cell strains, including metabolic enzymes and transcription factors. Of the clustered RNA species, only 36 were found to be altered at one time point in three or more of the cell strains analyzed (13 up-regulated, 23 down-regulated). Cluster analysis examined the effects of OTQ on the cells with specific p53 polymorphisms. The two strains expressing the major variant of p53 had 83 common genes altered (35 increased, 48 decreased) at one or more time point by at least a 0.6 signal log ratio (SLR). The intermediate variant strains showed 105 common genes altered (80 increased, 25 decreased) in both strains. Conclusion Differential changes in expression of these genes may yield biomarkers that provide insight into inter-individual variation in cancer risk. Further, specific individual patterns of gene expression may help to determine more susceptible populations.	Background Oxythioquinox (Morestan™ or OTQ, Bayer Corp) is a prototypical pesticide that was first used in 1968 on crops such as apples, pears, cucumbers, and gherkins. However, its use was later confined to non-food crops, limiting exposure to nursery and greenhouse employees. OTQ is a member of the quinoxaline class of pesticides, which also includes chlorquinox and thioquinox. Principal agricultural use of OTQ was limited to the states of California, Washington, Florida, New York, Pennsylvania, Ohio and Michigan [ 1 ]. The use of OTQ in the United States was voluntarily cancelled in 1999 with stocks in use until 2001, although OTQ is still listed in many different regions for use as an insecticide [ 2 , 3 ]. Further, OTQ is still in use today in other areas of the world, including Australia and the Caribbean [ 4 , 5 ]. Scientifically acceptable toxicity studies of OTQ are sparse. Early in vivo studies in rats found alterations of a variety of metabolic enzymes following OTQ exposures, including alkaline phosphatase [ 6 ]. Although not directly analyzed, the results of this study combined with an earlier study [ 7 ] suggest a direct effect of OTQ on succinate dehydrogenase, or other enzymes with a thiol group. Further work by Carlson et al (1970) [ 7 ] showed that, although OTQ has a low acute toxicity, cumulative exposure to this pesticide is not well-tolerated by exposed animals, with the majority of damage found in the liver of these animals. Further studies looking mainly at hepatic enzyme function found that OTQ exposure inhibits some hepatic enzyme functions [ 8 ]. OTQ was shown to be a carcinogen and hepatotoxin in laboratory animals in later studies [ 9 ]. OTQ has also been classified as a probable human carcinogen [ 10 ]. However, potentially carcinogenic exposures have already occurred and OTQ is still in use outside of the US, its mechanism of action remains of interest. In addition, OTQ has been shown to have an inhibitory effect on cytochome P450s, enzymes known to have a pivotal role in carcinogen metabolism [ 8 , 11 - 13 ]. Although early studies focused on hepatic effects, carcinogenic potential may occur in other tissues as well. While there have been multiple pesticides implicated in breast cancer, no studies have been published related to OTQ exposure and human cancer incidence [ 13 ]. This study is similar to work currently being carried out to determine gene expression profiles of a variety of environmental agents, including chemicals, physical agents and physiologic stresses [ 14 - 19 ]. The National Institute of Environmental Health Systems (NIEHS) has recently funded a consortium to expand the study of gene expression profile signatures for various chemicals, with the long-term goal to use these analyses in a validated gene expression profile signature database. A recent report by Shan etal. 2002 [ 20 ] is a good example of the differences between gene expression profile signatures for two chemicals in an animal system. This report profiled gene expression in rat carcinomas induced by two carcinogens, 2-amino-1-methyl-6-phenylimidazo [4,5- b ]pyridine (PhIP) and 7,12-dimethylbenz [ a ]anthacene (DMBA), and was able to show that while both chemicals altered expression in some genes in a similar fashion, each induced unique gene expression patterns. The primary goal of this study was to look for consistent changes common to all donors that could potentially be used as biomarkers of exposure, creating a gene expression profile following OTQ exposure in normal human cells. Given previous research in hepatoxicity, ideally normal human liver cells would have been used as a model system. However, we needed a normal human tissue that was readily accessible. Therefore, gene expression was studied in primary normal human mammary epithelial cells from four different donors in response to OTQ exposure. Current microarray analysis of pesticide exposure focuses on animal studies, not allowing for analysis of inter-individual variation. With the use of microarrays including clinical diagnosis, genetic susceptibility to disease and treatment, the need to determine the potential role of inter-individual variation on gene expression patterns in all potentially exposed tissues. Genes found to be altered in multiple cell strains could be considered biomarkers for individual populations. Given the small number of cell strains used, follow-up analysis in a larger number of samples is required to confirm any potential biomarkers. Biomarkers derived from this study could potentially be used in future epidemiology studies analyzing the effects of pesticide exposure. Further, this gene expression profile could be compared to those of other pesticides and of known carcinogens to develop a more detailed mechanistic definition of these chemicals. Methods Cell culture Primary normal human mammary epithelial cells (NHMECs) were derived from tissues salvaged at reduction mammoplasty obtained though the Cooperative Human Tissue Network (National Cancer Institute and National Disease Research Interchange). Development and characterization of cell strains was achieved using standard methods [ 21 ]. Cells were grown in MEGM media (Clonetics, Cambrex, Pittsburgh, PA) at 37°C and 5% CO 2 . OTQ treatment Treatment was performed on cells in passage six at 70% confluency, as routinely performed in our laboratory. Preliminary studies analyzed a range of OTQ concentrations (0 – 12.5 μM) selected based on previous research [ 6 , 7 ] and time points (0 – 24 hours), and showed maximum effect of OTQ on p53 expression with minimal toxicity at 2 hours with a final concentration of 6.25 μM. Cells were treated by diluting the stock OTQ/DMSO mixture in media and adding this solution to aspirated cells, allowing even exposure to all cells. DMSO (0.001%) alone was used as a vehicle control. At the end of the treatment period, cells were removed for RNA isolation. Cell viability was determined by Trypan Blue exclusion assay. Indirect immunofluorescence Confluent cells in passage five were trypsinized and plated on eight-well slides (LabTek II Slide System, Nunc, Naperville, IL). These cells were grown to 70% confluency before being treated with OTQ (6.25 μM) for 15, 60 and 120 min or DMSO (0.001%) for 120 min. At the end of treatment, the media was aspirated and the cells were fixed with methanol. Slides were then stained with anti- p53 antibody (1:1000, DO-1, Santa Cruz Biotech, Santa Cruz, CA) and incubated overnight at 4°C. The next day, the media was again removed and the secondary antibody (1:1500, goat anti-mouse FITC, Santa Cruz Biotech) was incubated for one hour at room temperature. Slides were washed in triplicate with phosphate buffered saline (PBS) and cover slips were added. Slides were dried for one hour at room temperature before viewing, using the laser scanning confocal microscope BX50 (Olympus), and quantitative analysis was performed [ 22 ]. Relative p53 expression was quantified between different time points and strains by determination of the area under the integrated intensity curve (Fluoview, Olympus, B & B Microscopes, Pittsburgh, PA). Microarray analysis Microarray analysis was performed in duplicate using the HuGeneFL high-density oligonucleotide microarrays (Affymetrix™, Santa Clara, CA). Protocols from the Affymetrix Expression Analysis Technical Manual were followed RNA was isolated from cells with Trizol (Gibco, Grand Island, NY), followed by purification with RNEasy Mini Kit (Qiagen, Valencia, CA). Spectrophotometer measurements were required to give a 260/280 ratio of 1.9–2.1 for use in microarray analysis. Double-stranded cDNA was then synthesized from total RNA (Superscript Choice System, Invitrogen, Carlsbad, CA). An in vitro transcription (IVT) reaction (Enzo, Farmingdale, NY) was then performed to produce biotin-labelled cRNA from the cDNA. Excess biotinylated dUTPs were removed by RNEasy Mini Kit before being fragmented and added to a hybridization cocktail, including Eukaryotic Hybridization controls (Affymetrix), BSA and herring sperm DNA (Gibco, Grand Island, NY) and biotinylated anti-streptavidin antibody (Vector Laboratories, Burlingame, CA). Hybridization on microarrays was performed for 16 hours at 45°C in the Gene Chip Hybridization Oven with rocker (Affymetrix). Microarrays were washed and stained using the protocol, as described in the Affymetrix Manual, with the GeneChip Fluidics Station 400 (Affymetrix). Arrays were then scanned with the Affymetrix Scanner (Hewlett Packard, Palo Alto, CA). Expression profiles were analyzed using Microarray Suite 5.0, MicroDB 3.0 and Data Mining Tool 3.0 (Affymetrix). Affymetrix arrays are produced using multiple 25-mer oligonucleotides (11–20 per target gene). Each oligonucleotide is created to match the selected region of the target gene (perfect match, PM), while a similar oligonucleotide is created altered in the 13 th position to control for non-specific binding (mismatch, MM). Results are given in signal intensities with a p -value determined from perfect match/mismatch (PM/MM) intensities by Tukey's Biweight analysis. Each array was normalized to a scaling factor of 1500 to correct for array variation. All arrays for each cell strain were analyzed on the same day to minimize variation. Signal log ratio was determined by comparison of the signal intensities for the baseline (vehicle control) and the treatment array. This is computed using a one-step Tukey's Biweight method by taking a mean of the log ratios of probe pair intensities across the two arrays. This method helps to filter out differences due to different probe binding coefficients that may lead to false positives and/or negatives. A signal log ratio of zero represents no change in gene expression as a result of OTQ exposure. A signal log ratio of one is equivalent to a fold change of two between the treatment and control. The results described here are the average of both duplicates, with the average percent variability between duplicate arrays being 1.5% (the average difference found between duplicates, related to array to array variability as well as technical variability in processing the array). Only relative changes equal to or greater than 0.6 signal log ratio (SLR) were considered a significant change as a result of exposure. The biological significance of each change is determined with Wilcoxon's signed rank test with the Affymetrix software. Gene chip analysis was performed by self-organizing map (SOM) clustering, focusing on genes with a detection p value of 0.05 or less at one or more time points. Analysis was performed to comply with MIAME standards. Real-time polymerase chain reaction analysis (RT-PCR) cDNA synthesized from each sample as in the Affymetrix analysis (Invitrogen) was used in a one-step RT-PCR analysis reaction. Analysis was performed in duplicate on the ABI 7700 cycler, with the SYBR Green Master Mix (ABI) and samples were normalized using both 18S and GAPDH expression levels for each sample. Primers were designed using Primer Express ® (ABI) to yield unique fragments for each gene under study. Reactions were set up following recommended protocols using 100 pmol of each primer (Sigma-Genosys) and approximately 60 ng template per reaction. Reactions were performed in duplicate for each sample for 40 cycles (95°C/15 sec denaturing step; 60°C/1 min annealing/extension step). Fold change was determined based on average cycle threshold (C T ) values for all duplicates and converted to signal log ratio. Results Trypan blue exclusion test Trypan Blue was used to analyze toxicity by measuring cell viability for each cell strain for each treatment. The results showed a range of viability from 92–97% at all time points, except for the last time point in strain 3, which had a viability of only 65% at 120 min (results not shown). This decrease in viability at 120 min was not found to be directly correlated to p53 or p21 protein expression, or to any particular gene expression pattern. The dose of OTQ used was based on these findings. Indirect immunofluorescence Baseline p53 protein levels were visually compared to those after treatment in each cell strain. Integrated fluorescence intensities were measured on each optical slice of cells. The fluorescence was determined as the area under the curve in arbitrary units (AU) (Figure 1 ). This result was compared between time points for each cell strain. An increase was seen in p53 expression in all cell strains with increasing duration of an OTQ exposure. However, one strain (3) showed a 10-fold lower p53 expression at each time point. Figure 1 Quantitative analysis of immunofluorescence microscopy. Confocal microscopy analysis of p53 expression. Integrated intensity measures were obtained from Fluoview and graphed with GraphPad Prism (GraphPad Software, San Diego, CA) to determine area under the curve as a measure of comparative p53 protein expression for each treatment time point. Data is shown for each time point per cell strain on the ordinate and the intensity on the abscissa (arbitrary units, AU). There was an increase in p53 protein expression for all of the strains tested in direct correlation with OTQ exposure. In one intermediate strain (3), p53 expression was 10-fold lower at all time points. DNA microarray DNA microarray analysis found no change in p53 , despite the increase in p53 protein levels observed by immunofluorescence. Studies of benzo [a]pyrene exposure in our laboratory have also found similar results for p53 expression [ 23 ]. The effect of OTQ exposure on other cell cycle genes, however, was determined by DNA microarray analysis. Although inter-individual variation between donors in response to OTQ was evident, there were also some genes found to be increased consistently in all strains by microarray analysis (Table 1 ). Self-organizing map (SOM) clustering was used to group genes with similar patterns of alteration in each of the strains. SOM analysis was performed following filtering of the total genes on the array, limiting the SOM analysis to only genes found to be present on at least one array analyzed. Following suggested analysis with the Affymetrix system, the default settings of the Affymetrix software were selected, including selecting threshold filtering (min = 20, max = 20000), row variation filtering (max/min = 3 and max-min = 100), and row normalization (mean = 0, variance = 1). Working with a 3 × 3 analysis to obtain 9 clusters generally gave an optimal amount of different clusters with less than 100 genes per cluster. From the SOM clustering analysis, genes altered by a signal log ratio of ± 0.6 or greater were chosen for closer study. Table 1 Genes altered following oxythioquinox exposure. Table represents data mined from HuGeneFL microarrays (Affymetrix). All genes selected have a signal log ratio of ± 0.6 unless otherwise noted. Representative genes for each group were selected based on their function and are shown here. GenBank ID Name Peak Expression Level (SLR) Functional Class Genes increased in three or more strains (n = 13): U22028 CYP2A13 1.5 xenobiotic metabolism U20734 junB 3.42 transcription V01512 cfos 2.04 transcription S85655 prohibitin 0.75 cell proliferation S82240 RhoE GTPase 2.81 signal transduction M69043 MAD-3 mRNA encoding IkB-like activity 0.83 apoptosis M63573 cyclophilin 1.68 immune response U05861 Dihydrodiol dehydrogenase 2.52 xenobiotic metabolism Genes decreased in three or more strains (n = 23): L05624 MAP Kinase Kinase -0.67 signal transduction U18018 E1A enhancer -1.34 transcription X56681 junD -1.03 transcription X68836 S-adenosylmethionine synthetase -2.57 cell metabolism J04973 Cytochrome bc-1 -3.12 cell metabolism Genes altered in at least two of four cell strains (n = 189): X03484 raf oncogene 1.46 carcinogenesis M60974 growth arrest and DNA-damage-inducible protein (gadd45) 1.51 DNA damage M57731 Human gro-beta 1.85 immune response X66899 EWS 1.2 carcinogenesis Z29087* Cyclin D1 Promoter 1.03 cell cycle control L10910 Splicing Factor CC1.3 0.62 RNA processing M83667 NF-IL6 Permeability Factor 1.55 transcription M27281 Vascular Permeability Factor -1.09 cell proliferation L28010 HnRNP F protein -0.55 RNA processing U72649 BTF2 -1.86 carcinogenesis M19267 tropomyosin -1.16 cardiac M38258 retinoic acid receptor gamma 1 -0.94 cell metabolism U42031 Immunophilin -1.41 immune response U67122 ubiquitin-related SUMO-1 -1.03 protein metabolism X70340 Transforming growth factor alpha -0.57 cell proliferation M34458 Lamin B -1.33 cell proliferation *No accession number was used by Affymetrix. This accession number most closely matches the probe description and sequence. SOM clustering was used to show patterns of expression in each cell strain, and followed by further subclustering of those clusters of interest across all cell strains (Figure 2 ). This analysis was performed with Data Mining Tool 3.0 (Affymetrix). For comparison to earlier versions of the Affymetrix software, a signal log ratio of one is equal to a fold change of two. SOM clustering led to the selection of genes found to be altered, with the overall total of 215 genes (13 increased in 3 or more strains; 23 decreased in 3 or more strains). To be included, the genes must have a signal log ratio of ± 0.6, which is approximately a 1.5 fold change. Signal log ratio was used to linearize the data for ease of analysis. The full list of genes altered can be found at . Figure 2 SOM clustering. Self-organizing map clustering groups' genes by similar expression patterns. Each data point represents an OTQ treatment time point. Data points are plotted in order of cell strain with four points per strain (DMSO, 15 min, 60 min, and 120 min). Levels of expression are not given, as patterns of expression are based on relative expression levels. Panels 4 and 7 of this graph show genes increased at 15 minutes for one strain (3) follow similar pattern in second strain (4). Similar pattern is also seen in panels 6 and 8 of this graph. From the full list of genes that fit these criteria, selected genes were chosen due to their potential role in carcinogenesis, whether by cell cycle control, immune response or other specific functions. Functions were described as annotated by NetAffx [ 24 ]. These genes are listed in Table 1 . Some genes were selected based on their possible role in disease as a result of pesticide exposure. Table 1 contains 8 genes increased in at least three of the four cell strains analyzed by 1.5 fold, as compared to the vehicle control and a list of 5 genes decreased in three or more of the four strains analyzed. Included in this list are two genes involved in the metabolic activation of endogeneous chemicals, cytochrome P4502A13 (CYP2A13) and dihydrodiol dehydrogenase . Although expression was slightly increased at one time point, the temporal patterns were slightly different (Figures 3 and 4 ). This increase in all cell strains suggests a potential role for these genes in OTQ metabolism. These results are confirmed at these time points as well as at 12 h and 24 h by RT-PCR (Table 2 ). Genes also found to be increased in at least three of four cell strains analyzed include genes involved in transcription ( junB , cfos ), immune response ( cyclophilin ), and apoptosis ( MAD-3 ). These genes showed a consistent increase in expression following exposure to OTQ (Table 1 ). Genes that showed a consistent decrease in expression following exposure include signalling pathway genes ( MAP kinase kinase ), cell metabolism genes ( S-adenosylmethionine synthetase , cytochrome bc-1 ), and transcription factors ( E1A enhancer ). Figure 3 Expression pattern for CYP2A13. DNA microarray analysis of NHMEC strains. Analysis was performed as described on HuGeneFL high-density oligonucleotide microarrays (Affymetrix). Results are plotted as duration of exposure vs signal log ratio (SLR). Signal log ratio is a measure of comparative expression of the treatment vs. vehicle control (0.001% DMSO). Signal log ratio of one is equal to a fold change of two. A SLR of 0.6 (Fold Change ~1.5) was the arbitrary limit of our analysis. All genes given an Absent call by analysis software are shown with a SLR of zero. Asterisks indicate a statistically significant variation in expression from the control level as measured by Tukey's Biweight analysis. Figure 4 Expression pattern for dihydrodiol dehydrogenase. DNA Microarray analysis of NHMEC strains. Analysis was performed as described on HuGeneFL high-density oligonucleotide microarrays (Affymetrix). Results are plotted as duration of exposure vs signal log ratio (SLR). Signal log ratio is a measure of comparative expression of the treatment vs. vehicle control (0.001% DMSO). Signal log ratio of one is equal to a fold change of two. A SLR of 0.6 (Fold Change ~1.5) was the arbitrary limit of our analysis. All genes given an Absent call by analysis software are shown with a SLR of zero. Asterisks indicate a statistically significant variation in expression from the control level as measured by Tukey's Biweight analysis. Table 2 RT-PCR results. Table represents real-time PCR data from selected genes of interest. Time points used were 15 min, 120 min, 12 h and 24 h where shown. Results shown are results of replicate analysis with duplicate samples. Samples not analyzed represented by N/A. Strain Gene SLR 15 min SLR 120 min SLR 12 h SLR 24 h 1 BTF2 1.41 N/A 1.53 0.3 2 BTF2 -1.56 N/A -1.32 -1.74 3 BTF2 0.18 N/A -0.89 -16.61 4 BTF2 0.62 N/A -0.36 0.65 1 CYP2A13 5.58 6.01 3.84 10.6 2 CYP2A13 2.7 0.62 1.1 -0.6 3 CYP2A13 5.76 4.77 -0.6 -1.64 4 CYP2A13 5.45 1.1 0.33 -1.51 1 DDH 1.23 -0.67 1.75 1.47 2 DDH -0.47 -0.62 0.96 5.41 3 DDH 0.86 -3.47 2.01 -3.84 4 DDH 3.44 N/A 5.53 6.09 1 EWS -1.18 -1.84 0.69 -5.64 2 EWS -6.64 -0.4 1.32 1.54 3 EWS -0.42 -0.22 -1.79 -6.64 4 EWS N/A 2.92 0.78 0.86 1 GADD45 -0.09 -0.22 -1.32 -5.06 2 GADD45 -1.03 -1.69 -1.94 -1.32 3 GADD45 0.82 -2.32 -1.56 -4.32 4 GADD45 1.34 9.05 6.4 7.72 1 MAD-3 -0.71 N/A -0.15 -3.64 2 MAD-3 -1.89 N/A -1.06 -0.38 3 MAD-3 0.34 N/A -0.58 -6.64 4 MAD-3 4.4 N/A -4.38 0.51 1 PROHIBITIN 0.31 N/A -0.3 0.96 2 PROHIBITIN -0.6 N/A 0.3 -1.18 3 PROHIBITIN -1.06 N/A -1.94 -2.84 4 PROHIBITIN 2.3 N/A -1.09 0.92 1 RAF 0.01 -2.25 1.21 -5.64 2 RAF -1.51 -0.27 -1.43 -0.06 3 RAF 0.37 -3.84 -1.12 -7.67 4 RAF 1.33 7.57 4.8 7.22 1 RhoE 0.59 N/A -0.62 -0.71 2 RhoE 0.26 N/A -1.18 1.08 3 RhoE 1.55 N/A -0.94 -5.64 4 RhoE 2.35 N/A 1.06 2.05 Some genes of interest were variably altered by strain, examples of which are also listed in Table 1 . These include genes involved in carcinogenesis ( raf oncogene , GADD45 , EWS , Cyclin D1 ) as well as immune response ( immunophilin , gro-β ), cell proliferation ( TGFβ ), and RNA processing ( HnRNP F protein ). Real-time PCR Real-time PCR was used to confirm and extend results seen by microarray analysis for selected genes. Following the original microarray analysis, patterns of some genes appeared to be changing at the latest time point (120 min), so extended time points were selected (12 and 24 h) to see a more complete expression profile for these genes. Due to limited amount of cDNA, genes were analyzed at 15 min and/or 120 min, and then analyzed at 12 and 24 h by RT-PCR. Extended time points were selected to look at specific genes found altered at the earlier time points. These genes were selected due to their function and/or pattern of expression, and determining their expression pattern at later time points was of interest. Results are shown in Table 2 . In the majority of samples, RT-PCR confirmed data found by microarray analysis for the genes listed. Some discrepancies are also shown, however, in these cases it is believed that the primer sequence is more specific to the gene in question than the probe sequence used on the array. However, extended time points, in some cases, showed that the results of early time points did not always continue to extended exposures. The RT-PCR results for prohibitin , DDH , and CYP2A13 at 24 h showed a decrease in expression in some of the cell strains analyzed. Samples not available for RT-PCR analysis are listed as N/A in Table 2 . Discussion The purpose of this study was to determine if microarray analysis of four normal human mammary cell strains with a known haplotype could be used to find biomarkers related to either exposure in general or the specific haplotype in question. The use of only four cell strains was determined to have enough power to provide basic information to lead to further study if necessary. This study would be followed up for specific genes of interest in a larger number of cell strains, preferably bypassing the more expensive and time-consuming microarray analysis for RT-PCR only. DNA microarray analysis was used to profile the cellular response to OTQ, a quinoxaline pesticide. Analysis revealed genes with common response across the four human cell strains studied as well as inter-individual variation in response. Using only four normal human cell strains, our goal was to discover any distinctly altered genes in response to OTQ exposure. Future analysis with a larger number of cell strains will be used to follow-up this analysis on specific genes of interest. The majority of studies looking at gene expression profiles have used animal models, limiting any knowledge obtained to genetically similar organisms. Analysis with normal human cell strains, like those described here, will give more information on inter-individual variation in response to various chemicals. This information will yield clues to the metabolic pathways of the specific chemicals, and this increased knowledge will aid in determining potential hazards in the environment and the workplace. Given the large number of pesticides in use today, further examination of the effect of these chemicals on individuals is warranted. Following exposure to OTQ, NHMECs showed alterations in genes involved in a variety of functions. These included xenobiotic metabolism, transcription, and DNA synthesis. Genes altered as a result of OTQ exposure in all strains analyzed included transcription factors like junB and cfos (the AP1 complex). A number of genes involved in carcinogenesis were found altered after exposure to OTQ, both induced and down-regulated. For example, prohibitin expression is found to be up-regulated in most cell strains after OTQ exposure, with similar expression patterns associated with a decrease in cancer incidence [ 25 ]. Metabolism genes that were altered after OTQ exposure, like cytochome P4502A13 (CYP2A13) and dihydrodiol dehydrogenase , are involved in xenobiotic metabolism. It has been suggested that CYP2A13 is the main metabolic activator of 4-(methylnitrosamino)-1-(3-pyridyl)-1-butanone (NNK), a tobacco-specific nitrosamine [ 26 ]. The increase of CYP2A13 begins to wane at the final time point tested, suggesting this alteration to be somewhat transient. This is an unusual expression pattern for a cytochrome p450, as genes in this family tend to show a gradual increase in induction, and an equally gradual decrease in expression. Further analysis by RT-PCR showed that this p450 was increased at later time points (12 h and 24 h, Table 2 ). Another metabolic enzyme affected by OTQ exposure is dihydrodiol dehydrogenase. Dihydrodiol dehydrogenase is known to participate in activation of certain polycyclic aromatic hydrocarbons (PAHs), so its increase in the intermediate variant after OTQ exposure may result in an increase in PAH activation [ 27 , 28 ]. Alterations of genes like this may suggest an indirect role for OTQ in carcinogenesis. Of these two genes, only CYP2A13 seems specific to exposure to OTQ. DDH has been found to be increased following exposure to various chemicals, including malathion, di-n-butyl phthalate, and benzo [a]pyrene (Gwinn in preparation, 2004) [ 23 , 29 ]. Some of the early results at 15 min by microarray analysis may have been a consequence of the stress of exposure, regardless of the chemical. Extending analysis to later times determined whether results seen at these early time points were still valid after longer exposure to OTQ. RT-PCR results given in Table 2 show that in most cases, the extended time points showed a continued trend of expression (whether increased or decreased), except in some cell strains for DDH , prohibitin , and MAD-3 . These results show a reverse of the early expression patterns at the later time points. Real-time PCR is a more specific method of analysis, as it only interrogates one gene at a time with primers designed uniquely to that gene. Conflicts in results between the two methods can generally be attributed to cross-reaction between probes designed for similar genes on the array. Sequence differences between the probes on the array and those used in RT-PCR may also play role in these results. The RT-PCR primers were selected specifically for the gene in question, while the probes on the array may not have been. Genes with sequence homology but with altered patterns of expression may not have been differentiated in the array analysis, but would be with the RT-PCR analysis. Inter-individual variation as a result of genetic polymorphisms in genes of interest would focus on specific at-risk worker populations. For example, the four cell strains analyzed in this study have been genotyped for a variety of genes, in particular those involved in cell cycle control and xenobiotic metabolism. Two of the four cell strains selected for analysis are heterozygote for the minor variant haplotype of p53, a cell cycle control gene (cell strains 3 and 4). Although no biological mechanism for the role of this variant in carcinogenesis has been defined, several studies associating this haplotype with various cancers support such role [ 30 - 35 ]. Analysis of genes altered in just those strains expressing this variant, including three genes involved in cell cycle control: raf oncogene (X03484), cyclin D1 (Z29087), and BTF2 (U72649) may further support an association between OTQ exposure, p53 variant status and carcinogenesis [ 36 - 38 ]. Of these, p53 has been reported to increase GADD45 transcription in response to DNA damage, which is associated with an increase in cell cycle arrest and DNA damage repair, while increased levels of raf oncogene have been associated with lung carcinogenesis [ 39 - 41 ]. Given the small number of cell strains used, this analysis needs to be extended to additional cell strains to determine the role of the p53 variants in gene expression differences. Due to the expense of microarray analysis, this is performed in only a limited number of cell strains (4), and selected genes will be further analyzed with \|RT-PCR in a larger number of cell strains. Over 80 cell strains have been established in our laboratory to date, with half of these having been genotyped for p53 . However, given that this haplotype is found only in a limited portion of the population, this varied pattern of expression in key genes in cell cycle control may highlight a specific at-risk population. Searches for similar natural compounds to replace these potentially disruptive chemicals can also use gene expression profiles [ 42 ]. Profile comparisons to that of a natural pesticide may decrease the need for organophosphates. Comparison of OTQ's gene expression profile to that of well-defined chemicals, like benzo [a]pyrene, will yield important information about OTQ's role in both genotoxicity and potential carcinogenicity. A comparison between many expression profiles is needed to further define similarities and differences between this pesticide and known carcinogens and/or other pesticides. Conclusions The overall goal of this project was to create a gene expression profile for OTQ or related pesticide analogues with the hopes of finding genes to be used as potential biomarkers of exposure. This expression profile may also be used to determine the final role of OTQ in carcinogenesis by comparing it to profiles of known carcinogens. It is possible that the main effect of OTQ exposure is not on the direct alterations in many genes, but on alterations in genes potentially involved in carcinogenesis, among them the examples of CYP2A13 and dihydrodiol dehydrogenase . The results shown here do not suggest a direct role of OTQ in carcinogenesis. They do, however, suggest OTQ exposure leads to an increase in expression of genes that do play a direct role in carcinogen; metabolism (for example, exposure to carcinogens like NNK and benzo [a]pyrene along with exposure to OTQ may lead to an increased incidence of tobacco-related cancer) [ 27 , 43 ]. Discovery of genes altered following exposure to OTQ in human cell strains may aid in future epidemiology studies on pesticide exposures. Gene expression profiling can be used to yield genetic biomarkers of exposure that, after validation, could be used in a clinical setting for early determination of organophosphate exposure, increasing early treatment of pesticide illness and thereby increasing the recovery rate of exposed individuals. List of abbreviations used OTQ, oxythioquinox DDH, dihydrodiol dehydrogenase DMSO, dimethylsulfoxide SLR, signal log ratio DMT, Data Mining Tool™ MAS, Microarray Suite™ SOM, self-organizing map NHMEC, normal human mammary epithelial cell RT-PCR, real-time polymerase chain reaction Competing interests None declared. Authors' contributions MRG participated in the design of the study, and performed all experiments. DLW was responsible for the growth and maintenance of all cell strains used. AW conceived of the study and participated in the design and coordination. All authors read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC521696.xml
548152	Upregulated expression of human neutrophil peptides 1, 2 and 3 (HNP 1-3) in colon cancer serum and tumours: a biomarker study	Background Molecular markers for localized colon tumours and for prognosis following therapy are needed. Proteomics research is currently producing numerous biomarker studies with clinical potential. We investigate the protein composition of plasma and of tumour extracts with the aim of identifying biomarkers for colon cancer. Methods By Surface Enhanced Laser Desorption/Ionisation – Time Of Flight / Mass spectrometry (SELDI-TOF/MS) we compare the protein profiles of colon cancer serum with serum from healthy individuals and the protein profiles of colon tumours with normal colon tissue. By size exclusion chromatography, we investigate the binding of HNP 1-3 to high mass plasma proteins. By microflow we investigate the effect of HNP 1-3 on mammalian cells. Results Human Neutrophil Peptides -1, -2 and -3 (HNP 1-3), also known as alfa-defensin-1, -2 and -3, are present in elevated concentrations in serum from colon cancer patients and in protein extracts from colon tumours. A fraction of HNP 1-3 in serum is bound to unidentified high mass plasma proteins. HNP 1-3 purified from colon tumours are lethal to mammalian cells. Conclusions HNP 1-3 may serve as blood markers for colon cancer in combination with other diagnostic tools. We propose that HNP 1-3 are carried into the bloodstream by attaching to high mass plasma proteins in the tumour microenvironment. We discuss the effect of HNP 1-3 on tumour progression.	Background The diagnostic stage of colon cancer determines survival. Patients diagnosed with localized tumours have a 75% probability of 5 year survival, whereas patients diagnosed with distant metastases only have a 5–10 % probability of 5 year survival (reviewed in [ 1 ]). Recently a number of studies have been published in which Surface Enhanced Laser Desorption/Ionisation-Time Of Flight/Mass Spectrometry (SELDI-TOF/MS) has been applied to biological samples from patients with various forms of cancer [ 2 - 4 ] leading to the identification of protein markers with clinical potential. Here we present a SELDI-TOF/MS study of colon cancer serum and tumours. We show that the expression of Human Neutrophil Peptides -1, -2 and -3 (HNP 1 -3), also known as alfa-defensin-1, -2 and -3, is upregulated in the tumour microenvironment, as compared to normal colon tissue. This finding is reflected in serum. We find that HNP 1-3 is present in elevated concentrations in serum from patients diagnosed with tumours in the colon, as compared to serum from a healthy control group matched by age and gender. By size-exclusion analysis we add to the existing evidence that HNP 1-3 bind to high mass plasma proteins, explaining the presence of HNP 1-3 in serum. By microflow analysis, we show that HNP 1-3 purified from colon tumours are lethal to mammalian cells. The HNP 1-3 peptides are part of the defensin family of peptides (reviewed in [ 5 - 7 ]), which are a fundamental component of the immune system and have the capacity to kill / inactivate a broad range of pathogens. Defensins are also known to function as regulators of both the innate and the adaptive immune system. We discuss the possible effects of HNP 1-3 in the tumour microenvironment. Methods Biological samples All biological samples were obtained by trained staff at Glostrup Hospital, Denmark. Written consent was obtained from all donators. Permission was obtained from the Danish Scientific Ethical Committee and the Danish Data Protection Agency. Tissue screening Normal colon tissue samples and colon tumour samples were obtained from the removed fragment of the patient's colon after surgical treatment for colon cancer. Tissue samples were stored at -80°C until use. The protein content was extracted from the tissue: 100 mg tissue sample was thawed on ice and homogenised on a Wheaton Overhead Stirrer for 2 minutes at speed step 2, in 500 ul Lysis buffer (100 mM TRIS-HCl, pH 8.0, 9.5 M UREA, 2% CHAPS). The samples were centrifuged at 14,000 rpm for 10 minutes and the pellet was discarded (repeated twice). The tissue protein extracts were stored at -80°C until use. Pilot studies were performed on different chips (data not shown) and the NP20 (Normal Phase) (Ciphergen) chip was chosen for the tissue screening. NP20 chips were placed in Bioprocessor (Ciphergen) and pre-treated with 50 ul tissue binding buffer (50 mM TRIS-HCl, pH 8.0) for 5 minutes on shaker (250 rpm) (repeated twice). 5 ul tissue protein extract was diluted in 50 ul tissue binding buffer and incubated in Bioprocessor on NP20 chips for 40 minutes at room temperature on shaker (250 rpm). Spots were washed twice in 250 ul tissue washing buffer (50 mM TRIS-HCl, pH 8.0) for 5 minutes. The chips were air dried for 10 minutes, followed by treatment with two times 0.6 ul 100% sinapinic (SPA) matrix solution. SPA was obtained from Ciphergen in 5 mg aliquots and dissolved (150 ul MQ water, 150 ul acetonitrile, 1.5 ul Tri-Fluoro-acetic-Acid (TFA)) immediately before the screenings. Serum screening Colon cancer serum samples were obtained from patients before surgical treatment. Normal serum was obtained from a group of healthy individuals matched by age and gender to the cancer patients. All serum samples were stored at -80°C until use. Serum pilot studies were performed on different chips to monitor the presence of HNP 1-3 in serum (data not shown). The immobilised metal affinity capture (IMAC30) chip was chosen for the actual screening and was pre-treated with nickel before analysis: 5 ul 100 mM NiSO4 were added to each spot and left on shaker (250 rpm) for 5 minutes (repeated twice). The chips were placed in Bioprocessor and incubated with 100 ul MQ for 5 minutes on shaker (250 rpm). Each spot was treated with 50 ul serum binding buffer (100 mM TRIS-HCl, pH 7.5, 500 mM NaCl, 0,1% Triton X-100) and left on shaker for 5 minutes (250 rpm). Serum samples were thawed on ice and 1 ul serum was diluted in 50 ul serum binding buffer and applied to spots and left on shaker (250 rpm) at room temperature for 40 minutes. The sample solution was removed and the spots were washed twice in 200 ul serum washing buffer (100 mM PBS, pH 7.4, 700 mM NaCl), followed by one wash in 200 ul MQ water. The chips were removed from the Bioprocessor and left to air dry for 20 minutes followed by treatment with two times 0.6 ul SPA. Only freshly made matrix solutions were used and the instrument was calibrated daily. Cancer and normal samples were run side by side. The chips were analysed on a PBS II instrument (Ciphergen). All spectra in each screening were normalised based on total ion current. Purification and identification of HNP 1-3 100 ul protein extract from colon tumour tissue in tissue lysis buffer was loaded unto a RP-HPLC column (uRPC C2/C18 ST 4.6/100, Pharmacia Biotech, Flow rate: 0.5 ml/min, Fraction size: 0.5 ml) in buffer A (0.065% TFA in MQ-water) and proteins were eluted in a gradient of 0–100% buffer B (0.05% TFA in acetonitrile (ACN)). Elution of peptides was monitored by absorbance at 280 nm. All protein-containing fractions were analysed by Matrix Assisted Laser Desorption/Ionisation-Time of flight (MALDI-TOF) on the PBS II instrument: 1.5 ul fraction was incubated with 0.6 ul SPA on a Gold array (Ciphergen) and left to crystallise, followed by an additional treatment with 0.6 ul SPA and analysed in the PBSII instrument. The HNP 1-3 containing fraction (32% buffer B) was further purified on a peptide gelfiltration column (Superdex Peptide HR 10/30, Pharmacia Biotech, Flow rate 0.9 ml/min, Fraction size: 0.5 ml, Buffer: 50% ACN, 0.1 % TFA). Elution of peptides was monitored by absorbance at 280 nm and protein-containing fractions were again analysed by MALDI-TOF. Purified HNP 1-3 were identified by on-chip trypsin digestion: 10 ul of HNP 1-3 fraction was applied to an NP20 chip and left on shaker (250 rpm) at room temperature for 40 minutes. The solution was removed and the spot was washed twice with 10 ul water. In order to denature peptides prior to digestion, the chip was left on heating block (80°C) for 5 minutes. The chip was cooled on ice for 2 minutes. 10 ul trypsin digestion solution (0.01 ug/ul trypsin in 50 mM NH 4 HCO 3 , pH 8,0) was added, and the chip was left for 10 hours at 40°C in humidity chamber after which the chip was left to air dry for 20 minutes. 1 ul CHCA matrix (prepared as the SPA matrix solution) was added and the peptide map was analysed on the PBS II instrument. Peptide maps of trypsin autodiggest were used as controls. Identification was done with the PepIdent software on the Expasy server. For the reduction experiment, HNP 1-3 were first denaturation by heating (10 minutes at 80°C) followed by treatment with DTT (200 mM, 30 minutes at room temperature) and the peptides were incubated on an NP20 chip and analysed on the PBS II instrument. Size exclusion chromatography of HNP 1-3 50 ul colon cancer serum was loaded unto a peptide gelfiltration column (Superdex Peptide HR 10/30, Pharmacia Biotech, optimal separation range: 1 to 7 kDa, flow rate: 0.5 ml/min, fraction size: 0.5 ml, buffer: 10 mM Ammonium carbonate, pH: 8.0). Elution of peptides was followed by absorbance at 280 nm. All protein containing fractions were analysed by MALDI-TOF on PBS II (Ciphergen) as described above. Maximum signal intensity of 40 individual peaks was plotted as a function of elution volume and an approximate elution curve was calculated. Study of HNP 1-3 by microflow For micro flow experiments, canine kidney cells (MDCK cells) were plated onto poly-d-lysine coated cover slips at a concentration 3000 cells/well, grown in Dilbeccoo's Modified Eagle Medium (DMEM) with 10% Fetal Bovine Serum (FBS) for five days with the result of confluent islands. Microflow was performed in an Eppendorf micromanipulator 5171 and transjector 5246 system mounted on a Leica DMIRBE inverted research microscope. Micro capillaries (borosilicate with filament, Sutter Instruments Company, Novato, California, USA) were pulled to an outer diameter of 0.85 nm on a Sutter P-97 Micropipette Puller. The dye-loaded cells were visualized by excitation at 470 nm and recorded at 509-nm emission using Haupage version 3.3.18038 software and Kappa CF 15/4 MC-S camera (Leica). The MDCK cells were recorded (in CO2 independent media) on the inverted DMIRBE inverted research microscope. The capillary was placed 20 nmover the confluent cells with a constant flow (1300 hPa). The MDCK cells were exposed to peptide and calcein (20 mM) fractions for 60 minutes. Results HNP 1-3 expression in tissue and serum We performed pilot studies of colon tumour and normal colon tissue on a variety of chips with different chemical properties and with different binding and washing conditions. Based on these preliminary studies, we found that the expression of three peptides with mass/charge ratio (m/z) values of 3372, 3443 and 3486 (subsequently identified as HNP 2, 1 and 3, respectively), were upregulated in the tumour samples. The three peptides were visible on different chips and under different binding conditions (data not shown). The strongest signal of HNP 1-3 in tissue extract was obtained on the NP20 (Normal Phase) chip, whereas the strongest signal of HNP 1-3 in serum was observed on the IMAC30 (Ni) (immobilised metal affinity capture chip, activated with nickel), and these conditions were chosen for the actual screenings. We emphasize that in general the protein profiles of serum and tissue were very different when using the same protein chip for serum and tissue extract. However, individual peaks were present on several types of chips, and observed in both serum and in tissue extract, for example the characteristic triplet with m/z 3372, 3443 and 3486. Protein extract from 40 colon tumour and 40 normal colon tissue samples were analysed on NP20 chips and 125 colon cancer serum samples and 100 normal serum samples were analysed on IMAC30 (Ni) chips. All spectres in each screening were pooled and normalised based on overall ion current. Each spectrum produced approximately 40 to 90 protein peaks in the range from 2 to 80 kDa (FIG. 1A–C ). Statistical analysis of the intensity values of HNP 1-3 in the tissue screening (FIG. 2A .) showed that HNP 1-3 were significantly upregulated in tumours (p < 0.0005) and statistical analysis of HNP 1-3 expression in the serum screening (FIG. 2B .) showed that HNP 1-3 were significantly upregulated in cancer serum (p < 2.2e -16 ). Compared to other peptides in the same range, HNP 1-3 showed average signal intensity in most normal colon tissue extract, whereas the HNP 1-3 signal was extremely high in most tumour samples (in some tumour spectres the HNP 1-3 peaks were the strongest of all peaks). In the normal serum samples the HNP 1-3 signals were weak and only slightly stronger in the cancer serum. Identification of HNP 1-3 The markers were purified by RP-HPLC, peptide gelfiltration and on-chip purification, after which they were identified by peptide mapping as HNP-2 (3372 Da), HNP-1 (3442 Da) and HNP-3 (3486 Da) (Table 1A ). The masses correspond to the peptides in their oxidised states with three disulfide bridges. After reduction with DTT, HNP-1 and HNP-2 increased 6 Daltons in mass, due to reduction of the six cysteines (Table 1B ). We were not able to reduce HNP-3. Size exclusion chromatography of HNP 1-3 50 ul colon tumour extract in Lysis buffer was applied to a peptide gelfiltration column. Elution of peptides was followed by absorbance at 280 nm. All fractions were analysed by MALDI-TOF on the PBS II instrument (Ciphergen). Maximum signal intensity of 40 individual peaks was plotted as a function of the elution volume and an approximate elution curve was calculated (FIG. 3 ). HNP 1-3 peptides were primarily eluted in early fraction together with high mass proteins above 20 kDa and also, but to a lesser degree, in fractions together with other peptides of similar mass interval (2 to 4 kDa) (FIG. 3 ). Cytoxic assay The cytotoxicity of HNP 1-3 purified from colon tumours was tested by exposing MDCK cells to different fractions purified from colon tumours. Calcein were added to the fractions and the solutions were left to flow over the cells for one hour. By fluorescence microscopy calcein was observed to accumulate in cells exposed to HNP 1-3/calcein fractions, whereas cells treated with fractions containing other (unidentified) tissue peptides did not uptake calcein (FIG. 4C&D ). Further, by microscopy we observed that cells exposed to HNP 1-3 appeared more diffuse and had enlarged nuclei, indicating apoptosis (FIG. 4A&B ). Discussion Elevated concentrations of HNP 1-3 in colon cancer serum Abnormal concentration of HNP 1-3 in blood has previously been demonstrated in connection with benign conditions. Elevated concentrations of HNP 1-3 following infection (bacterial- / non-bacterial- infection and pulmonary tuberculosis) has been found in plasma, blood and other body fluids [ 8 ], and plasma HNP 1-3 concentrations have been shown to be elevated in patients with septicemia or bacterial meningitis [ 9 ]. HNP 1-3 have been found in urine from patients with transitional cell carcinoma of the bladder [ 10 ] and HNP-1 has been found in salvia of patients with oral carcinomas [ 11 ]. Finally, HNP 1-3 are found in excess amounts in tears after ocular surface surgery [ 12 ]. Our study is the first that demonstrate elevated concentrations of HNP 1-3 in blood following tumour growth. Elevated concentrations of HNP 1-3 in colon tumours HNP expression has previously been linked to different types of tumours and cell lines. HNP-1 has been detected in lung tumours [ 13 ] and in the submandibular glands of patients with oral carcinomas [ 14 ]. By RT-PCR, mass spectrometry and flow cytometric analysis, HNP 1-3 have been shown to be expressed by cell lines deriving from renal cell carcinomas [ 15 ] and the expression of a specific HNP precursor peptide has been shown to be upregulated in human leukemic cells [ 16 ]. Our results suggest that HNP 1-3 are extremely abundant in colon tumours. This is in agreement with a study of HNP-1 in lung tumours, where the maximum observed level was 26 nanomoles per gram wet tissue [ 13 ]. In order for these excessive amounts of peptide to be detectable in serum, the peptides must be present extracellular in the tumour environment, such as observed in studies of HNP 1-3 expression in kidney [ 14 ] and brain [ 17 ]. Usually tumour expressed peptides are not easily detected in serum or plasma by SELDI-TOF mass spectrometry techniques; the highly concentrated plasma proteins out compete the signal of low abundance peptides. We suggest that HNP 1-3 are detectable in serum only because they are expressed in exceptionally high amounts in the tumour microenvironment. Previous studies indicate that HNP 1-3 expression in tumours primarily originate from tumour invading eosinophils [ 13 ] and neutrophils [ 14 , 18 ]. However, it has been shown that the excess amounts of HNP 1-3 observed in urine from bladder cancer patients were often produced by the actual bladder cancer cells [ 10 ], and that highly invasive bladder cancer cells produced more HNP 1-3 than less invasive ones (Holterman DA, in press, personal communication). Since our tissue screening is based on comparison of tissue samples, we cannot say whether the peptides are produced by the colon cancer cells or by tumour infiltrating neutrophils. In the case of active inflammatory bowel disease, it is not clear whether epithelial expression of HNP1-3 is induced by the inflammatory state or the whether the peptides are released by adjacent neutrophils and taken up by the epithelial cells [ 6 ]. Here it is believed that the peptides provide protection against microbial invasion when the mucosal barrier is damaged (such as is the case in inflammatory bowel diseases). HNP 1-3 are known to stimulate bronchial epithelial cells to upregulate interleukin-8 production [ 19 ], a potent neutrophil chemotactic factor. Thus, the upregulated expression of HNP 1-3 in tumours may primarily originate from invading immune cells, but could be initiated by HNP 1-3 producing cancer cells. Size exclusion chromatography of HNP 1-3 We explain the elevated concentrations of HNP 1-3 in colon cancer serum by unspecific binding between HNP 1-3 and high mass plasma proteins. We suggest that the peptides attach to plasma proteins in the tumour area and are carried into the bloodstream. The HNP 1-3 we observe in high mass fractions from size exclusion, could also be explained by multimerisation: In one study it was demonstrated that defensins form voltage dependent channels in lipid bilayer membranes and further conductance investigations suggested that the channels were formed by multimers containing 2–4 molecules [ 20 ] and a crystal structure study [ 21 ] of HNP-3 revealed an amphiphilic dimmer. We interpret the size exclusion results as evidence for interaction between HNP 1-3 and unidentified high mass proteins through unspecific interactions; the peptides are eluted in very early fractions which would probably require the presence of multimers of more than five peptides and our study does not reveal the presence of multimers containing four, three or two peptides. Further, our interpretation is in agreement with a number of previous studies that show that HNPs are bound to plasma protein in vitro and that high concentrations of HNPs causes precipitation of plasma proteins; specifically 2-macroglubulin and C1 complement [ 22 , 23 ] has been shown to bind defensin. Another study [ 24 ] showed that HNP-1 bind to various plasma proteins, notably serum albumin, and it was found that serum, or serum albumin, was able to inhibit the anti-viral activity of HNP-1. This ability to bind to plasma proteins could also explain why HNP 1-3 lysis of mammalian cells is hindered in the presence of serum [ 25 ]. Together these observations add to the growing realisation that common plasma proteins may carry disease specific peptides and therefore should not be ignored in biomarker research. Common to beta-defensin 2 (another member of the defensin family) and HNP 1-3 is an uneven distribution of surface charges. Beta-defensin 2 has been shown to bind to a chemokine receptor [ 26 ]. It has been suggested that the positively charged cluster found in defensin peptides and chemokines, may play a common role in binding to receptors, but is not important for determining receptor specificity [ 27 ]. This surface charge may also explain the binding of HNP 1-3 to plasma proteins. The observation that defensins are localised to lymphocyte nuclei [ 28 ], could similarly be explained by unspecific binding to shuttle proteins. The function of HNP 1-3 The exact concentration of HNPs in the tumour microenvironment may decide the in vivo function of HNP 1-3. One study showed that HNP 1-3 mediates lysis of tumours in a concentration dependent manner [ 25 ]. This is in agreement with another study that show that only relatively high concentrations of HNP-1 (10 -4 M) are cytotoxic for human monocytes, whereas lower concentration of HNP-1 (10 -8 to 10 -9 M) increases TNF-alpha production by monocytes [ 29 ]. In a study of renal cell carcinoma lines [ 14 ] it was shown that HNP 1-3 were cytotoxic to all tested cell lines when present in high concentrations (above 25 ug/ml), but at lower concentration HNP 1-3 stimulated growth of a subset of tumour cell lines. We add to these results by demonstrating that in vivo HNP 1-3 purified from colon tumours are capable of lysing MDCK cells. Our study was based on a 60 minutes microflow study. This screening set-up did not allow us to investigate the minimum concentration of HNP 1-3 necessary for lysis. Conclusions The high concentration of HNP 1-3 observed in tumours and the observation that HNP 1-3 are capable of lysing mammalian cells may lead to the conclusion that the peptides serve to the benefit of the host by killing tumour cells. However, in one study HNP 1-3 were found to bind to HLA-Class II molecules and were capable of reducing the proliferation of a HLA-DR-restricted T-cell line after stimulation [ 30 ] and could in this way help the cancer cells avoid local immune recognition. Defensins also regulate the systemic immune response. Through interaction with the chemokine receptor CCR6, beta-defensins recruits dendritic and T cells[ 26 ] (discussed in [ 31 ]) and HNP 1-3 are capable of recruiting leukocytes to sites of infection in mice [ 32 ]. Upregulated immune responses are known to stimulate tumour proliferation; immune cells are actively recruited by tumours to exploit their pro-angiogenic and pro-metastatic effects (reviewed in [ 33 , 34 ]). Whether the high concentrations of HNP 1-3 in the tumour limits the tumour growth or on the contrary stimulate tumour proliferation is not clarified; we emphasise that HNP 1-3 are expressed in inflammatory bowel disease and could be involved in the early stages of carcinogenesis. We suggest that the prominent surface charge on defensins, their unspecific binding to other proteins (such as high mass plasma proteins) and the observed excess amounts of peptides found in tumours, could provide the peptides with broad antagonising effects on numerous receptors in the tumour microenvironment. In this way HNP 1-3 peptides may serve to the benefit of the tumour. Our results add to the evidence that HNP 1-3 bind to high mass plasma proteins. We suggest that the peptides are released in the inflammatory site and passively diffuse into the blood stream. HNP 1-3 have been observed in primary tumours of different tissues and the peptides are known to play a fundamental role in the innate immune system. We suggest that the peptides may serve as markers for colon cancer in combination with established diagnostic tools and as prognostic markers following therapy. Competing interests This study was partly financed by Colotech ltd. and partly by Glostrup Hospital. Jakob Albrethsen, Rikke Bøgebo and Hans Raskov receive salary from Colotech A/S. Steen Gammeltoft, Jesper Olsen and Benny Winther receive salary from Glostrup Hospital. Authors' contributions JA performed the SELDI-TOF/MS screenings, the protein identification experiments and the size exclusion study. RB did the statistical analysis, BW did the microflow study, JO obtained the biological samples. HR and SG planned the project. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC548152.xml
546037	Light-Dependent Development of Circadian Gene Expression in Transgenic Zebrafish	The roles of environmental stimuli in initiation and synchronization of circadian oscillation during development appear to vary among different rhythmic processes. In zebrafish, a variety of rhythms emerge in larvae only after exposure to light-dark (LD) cycles, whereas zebrafish period3 (per3) mRNA has been reported to be rhythmic from day 1 of development in constant conditions. We generated transgenic zebrafish in which expression of the firefly luciferase (luc) gene is driven by the zebrafish per3 promoter. Live larvae from these lines are rhythmically bioluminescent, providing the first vertebrate system for high-throughput measurement of circadian gene expression in vivo. Circadian rhythmicity in constant conditions was observed only after 5–6 d of development, and only if the fish were exposed to LD signals after day 4. Regardless of light exposure, a novel developmental profile was observed, with low expression during the first few days and a rapid increase when active swimming begins. Ambient temperature affected the developmental profile and overall levels of per3 and luc mRNA, as well as the critical days in which LD cycles were needed for robust bioluminescence rhythms. In summary, per3-luc zebrafish has revealed complex interactions among developmental events, light, and temperature in the expression of a clock gene.	Introduction The circadian clock controls biological processes such as behavior, gene expression, and physiology in diverse organisms, ensuring that these processes to take place at appropriate times of the day. This is crucial for many organisms, such as plants, which must synchronize photosynthesis with day-night cycles. Animals also synchronize to environmental cycles because of more subtle but nevertheless important needs such as predator avoidance, food availability, and optimal temperatures for various processes. Circadian clocks in all species share the following properties: They persist even in constant environmental conditions with periods near 24 h; they can be reset by environmental stimuli such as light and temperature; and their periods are relatively constant at different temperatures. Recently, remarkable progress has been made in elucidation of molecular mechanisms of circadian clocks in diverse organisms. The common feature of clocks is cycling gene expression due to intracellular transcriptional feedback loops [ 1 , 2 , 3 ]. Genetic analysis in higher metazoan species, for example the fruit fly Drosophila, has been extremely valuable in identifying important players of the clockworks and their roles [ 4 , 5 , 6 ]. Identification of clock genes in Drosophila lead to molecular dissection of clock mechanisms in mammals mainly by testing whether homologs of Drosophila clock genes are involved in mammalian clocks. While this approach is informative, it harbors the risk of missing important factors that would have been found by forward genetic searches without preconceptions. Furthermore, gaps in our understanding of the clock mechanisms include factors responsible for the expression of positive transcription factors Clock and Bmal and mechanisms for clock protein turnover. In this regard, zebrafish is in a unique position as a vertebrate species in which large-scale forward genetic screens are convenient [ 7 , 8 ]. Furthermore, zebrafish circadian clocks have been shown to possess unique properties such as light entrainability of molecular rhythms in cultured organs and cells [ 9 , 10 , 11 ]. A behavioral screening for circadian mutations has been successfully carried out in zebrafish [ 12 ]. However, due to the limited capacity of this method, it is not suited to high-throughput screening. A method relying on bioluminescence rhythms mediated by luciferase reporting has been successfully used to screen for mutants affecting circadian gene expression in plants, cyanobacteria, and flies [ 13 , 14 , 15 ]. In vertebrates, however, luciferase reporting has been used mainly for recording circadian gene expression in cultured tissues and cells, because of technical difficulties in these species [ 16 , 17 , 18 , 19 , 20 ]. Bioluminescence rhythms mediated by a zper4-luc promoter fusion construct have been studied successfully in the zebrafish PAC-2 cell line [ 21 ]. While this approach was useful for promoter dissection, generation of transgenic animals is necessary for mutagenesis screening. In this study, transgenic zebrafish were made in which cycling expression of the firefly luc gene is driven by the promoter of per3 [ 22 ]. This promoter was chosen because per3 mRNA has been shown to oscillate rhythmically in embryos as well as in a cell line [ 11 , 22 ]. For mutagenesis screening, it is most convenient and economical to test the youngest possible animals and avoid raising the animals that give negative results. In this regard, per3-luc was considered ideal for mutagenesis screening, because an in situ hybridization study showed that per3 mRNA cycles starting on day 1 postfertilization with or without any entraining signals [ 22 ]. It was suggested that maternal per3 mRNA present in the oocyte can set the phase of per3 mRNA rhythms in early embryos. This result, however, is not consistent with other studies involving development of circadian rhythms: Rhythms of melatonin production require a light-dark (LD) transition later than 20 h postfertilization [ 23 ]; circadian swimming rhythms in larval fish develop during the first 4 d of development and require entraining signals late in embryonic development [ 24 ]; and rhythms of cell proliferation in larval fish develop only after exposure to several LD cycles [ 25 ]. In order to determine the earliest possible developmental time when rhythmic luc expression can be monitored in the per3-luc transgenic fish, embryos from the transgenic lines were monitored for bioluminescence from day 1 of development ( Protocol S1 ). To our surprise, very low and non-oscillating levels of bioluminescence were detected during the first 4–5 d into development. Furthermore, consistent with the rhythms of melatonin production, locomotor activity, and cell division, rhythmicity of the per3 gene expression gradually developed during the first several days postfertilization, and was observed only if fish were exposed to LD cycles during the hatching period or later. It was also found that ambient temperature affects per3-luc -mediated bioluminescence in a complex way. This study defines conditions under which the per3-luc transgenic fish can be used for mutagenesis screening and other types of studies. Results Generation of per3-luc Transgenic Fish To develop a system in which circadian gene expression in zebrafish can be monitored in vivo, transgenic fish were generated in which the expression of the firefly luc gene is driven by the promoter of the per3 gene [ 22 ]. The construct was made by modifying a bacterial artificial chromosome (BAC) originally screened for sequences in the first coding exon of per3 ( Figure 1 ) . By comparison of per3 cDNA sequence to the genomic sequence from another BAC clone (CH211–138E4) from this region, it was found that the cDNA contains another exon 5′ to the first coding exon. By comparing BAC-end sequences from the construct to genomic sequences from CH211–138E4 as well as with the Ensembl Zebrafish whole genome shotgun assembly sequence version 4 ( http://www.ensembl.org/Danio_rerio/ , the construct was found to be approximately 72 kb long, and spans from 26 kb upstream of exon 1 to intron 19 of per3, and contains part of another gene 5′ to per3 ( Figure 1 ). One canonical and two noncanonical E-boxes were found within 1 kb of the per3 promoter (unpublished data). In the modified BAC, the coding portion of the first coding exon was replaced by the luc and kanamycin resistance (Km r ) genes ( Figure 1 ). This rather long construct was made because it has been shown in zebrafish that a reporter gene is more consistently expressed in a context of a longer BAC construct than in a conventional short construct with just a few kilobases of promoter sequences [ 26 ]. Figure 1 Schematic Map of the per3-luc Construct The top graphic shows the exon-intron structures of per3 and the flanking gene. The BAC clone 8M06 screened for the first coding exon of per3 is approximately 72 kb long, and extends from about 26 kb upstream of exon 1 to intron 19 of per3 . The bottom graphic shows the magnified view of the first coding exon in the BAC 8M06 and the modified BAC construct. The white and black boxes represent noncoding and coding sequences, respectively, of the first coding exon of per3 . The coding sequence of this exon was replaced with an approximately 3-kb fragment containing luc and Km r . Arrows under the construct represent primers used for the screening of transgenic lines. After screening 147 injected founders by PCR, five independent transgenic lines were found. Each positive founder was bred to a wild-type fish, and their progeny were individually tested for bioluminescence as 5–7-d-old larval fish. Larval fish with bioluminescence above background (more than 100 counts per second [cps]) were raised as transgenic F1 fish. Three of the five lines emitted bioluminescence above background. The level of bioluminescence varied depending on the line. The strongest-glowing line (#23) was used for this study unless otherwise stated. All the animals used in this study were the progeny of crosses between a transgenic line and the AB wild-type strain. Therefore, these animals carried the transgene in hemizygous condition. Light Signals Are Necessary for per3-luc Rhythms One of the intended usages of the per3-luc transgenic zebrafish is screening for mutations that affect bioluminescence rhythms. Since it is most convenient to screen the youngest possible animals, bioluminescence from transgenic embryos was monitored first. It was also expected that per3-luc -mediated bioluminescence in embryos should cycle from day 1 of development even in constant conditions, because per3 mRNA expression detected by in situ hybridization has been demonstrated to oscillate from day 1 postfertilization in constant conditions [ 22 ]. Therefore, embryos carrying the transgene in hemizygous condition were collected and their bioluminescence monitored for 10 d starting from day 1 postfertilization. Surprisingly, when embryos were exposed to only one 14 h light: 10 h dark (14:10 LD, lights on at 8 A.M. ; lights off at 10 P.M. CST) cycle on day 1, the majority of the animals showed no circadian rhythmicity of bioluminescence ( Figure 2 A and 2 B; Table 1 ). Nevertheless, a characteristic developmental profile of luc expression was observed. Bioluminescence mediated by per3-luc stays rather low until day 4, when there is a small peak of bioluminescence, followed by a small dip on day 5 and a rapid increase that reaches the second peak on days 7–9. The slow decline of luminescence after that point may be due to substrate deprivation common in luciferase reporting [ 27 ]. Importantly, this developmental profile was also observed in two other lines of per3-luc, albeit with much lower overall luminescence counts (unpublished data). Figure 2 Bioluminescence in Embryos That Experienced Different Numbers of LD Cycles during Development Embryos hemizygous for per3-luc were collected and monitored for bioluminescence while exposed to different numbers of 14:10 LD cycles starting on day 1 postfertilization followed by DD. (A) One LD, (B) one LD, (C) two LD, (D) three LD, (E) four LD, (F) five LD, (G) six LD, and (H) six LD. In (I), embryos were exposed to two LDs, one on day 1 and the other on day 6 of development. Black and white bars on top of each plot represent the times when the lights were off and on, respectively. Since overall bioluminescence levels can vary among clutches and experiments, normalized bioluminescence was averaged and plotted in each graph. Number of animals that were averaged is given at top right corner of each plot. Error bars represent standard error of the mean (SEM) For the one-LD and six-LD groups, plots for two experiments are shown here. These experiments showed small differences in developmental profiles, possibly due to differences in room temperature (about 1 °C), therefore could not be pooled. For the two- to five-LD groups, data from two experiments were pooled. The small but abrupt increase of luminescence that occurred only during the light period of LD cycles is considered an artifact made visible by low bioluminescence counts during the first several days of development, because the same level of fluctuation was observed in empty wells under LD condition. Table 1 Rhythmicity and Periods of Larval Zebrafish: Varying Number of LD Cycles Data are included for 5.5- to 10-d-old larval fish that had experienced varying numbers of LD during development. For each group, results of two experiments were pooled. Mean ± SEM of rhythm statistic [ 54 ] and period were calculated for rhythmic individuals only. A series of G-tests showed that percent rhythmic values were significantly different ( p < 0.001) among the seven groups a Further tests showed that these percent rhythmic values are significantly higher ( p < 0.05) than the others. Except for the one- to three-LD groups that were mostly arrhythmic, differences in rhythm statistic among groups were not quite significant (Wilcoxon/Kruskal-Wallis test, p = 0.027, α = 0.017) Since the light signal on day 1 was not enough to elicit detectable rhythmicity of bioluminescence, an increasing number of LD cycles were given to embryos while they were monitored for bioluminescence (see Protocol S1 ). LD cycles on days 2 and 3 did not increase rhythmicity on subsequent days, although small fluctuations of bioluminescence were discernible on days 7–10 in the averaged plots ( Figure 2 C and 2 D; Table 1 ). The number of animals expressing significant rhythmicity during the last 4.5 d of the record increased gradually when the number of LD cycles was increased from three to six ( Figure 2 D– 2 H; Table 1 ). The circadian fluctuation was superimposed on the developmental profile also seen in embryos entrained by fewer numbers of LD cycles ( Figure 2 ). To determine whether the number of LD cycles or the developmental stage at which the last LD transition occurred is more important for robust rhythmicity of luciferase reporting, embryos were monitored for bioluminescence while experiencing two LD cycles, one on day 1 and the other on day 6 ( Protocol S1 ). Luminescence rhythms on the last 4.5 d of the record for this group of animals were as robust as those exposed to six LDs ( Figure 2 G– 2 I; Table 1 ). Thus, the developmental stage at which the last LD transition occurred, rather than the number of LD cycles, was important for light entrainment of per3-luc rhythms. There was no systematic effect of the time of the last lights-off on free-running periods ( Tables 1 and 2 ). Table 2 Rhythmicity and Periods of Larval Zebrafish: Varying LD and Temperature Data are included for 5.5- to 10-d-old larval fish that had experienced varying numbers of LD at different temperatures during development. Results from one of two experiments with similar results are shown here. Mean ± SEM of rhythm statistic and period were calculated only for rhythmic individuals. A series of G-tests showed that percent rhythmic values among the six groups were significantly different ( p < 0.001) a Further tests showed that these percent rhythmic values were significantly higher ( p < 0.01) than the others. For the six-LD groups, the rhythm statistic was not significantly different between the two temperature groups ( t -test, p = 0.61) Luciferase Reporting Reflects per3 Expression The lack of bioluminescence rhythms during the first few days of development was rather unexpected because of the previously reported per3 mRNA rhythms [ 22 ]. Therefore, mRNA cycling of per3 and luc was compared by real-time quantitative PCR (qPCR). As in the experiment shown in Figure 2 G and 2 H, embryos were exposed to six LD cycles and transferred to constant darkness (DD). Embryos were collected and their mRNA was extracted on days 3 and 8. Both per3 and luc mRNA levels were much lower on day 3 compared to day 8 ( Figure 3 ). What appears to be approximately 2-fold oscillations of per3 and luc mRNA on day 3 (see the insets on Figure 3 A and 3 C) were not statistically significant ( p = 0.47 for per3 , and p = 0.08 for luc by the Wilcoxon/Kruskal-Wallis test). In contrast, approximately 5-fold fluctuations of the transcripts were observed on day 8 ( Figure 3 B and 3 D). This increase in overall expression levels and cycling amplitudes reflect the observed bioluminescence profile, although there were qualitative differences between bioluminescence and RNA as well as between per3 and luc mRNA (see Discussion). Importantly, the peak phase of mRNA cycling on day 8 was 5–7 h advanced compared to the phase of bioluminescence cycling (compare Figure 2 G and 2 H to Figure 3 B and 3 D). Figure 3 Temporal Expression of luc mRNA Is Similar to That of per3 during Development Embryos hemizygous for per3-luc were collected from naturally breeding parents and kept in 14:10 LD cycles (lights on at 8 A.M. CST) at 22 °C for 6 d. Larval fish were shifted to DD at the end of the light phase on day 6. Total RNA was extracted from 2- to 3-d-old embryos and 7- to 8-d-old larval fish, and was subjected to real-time PCR for per3 and luc mRNA levels. (A) Expression of per3 mRNA per embryo on day 3 was determined every 6 h starting at 10 A.M. (2 h after lights-on). (B) The per3 mRNA per animal on day 8 was measured every 4 h starting at 10 A.M. (C) Levels of luc mRNA on day 3 were determined as in (A). White and black bars at the bottom represent light and dark phases, respectively, for (A) and (C). (D) Cycling of luc mRNA on day 8 was determined as in (B). The gray and black bars represent the time when the light would have been on and off, respectively, had the LD cycles continued. For each of per3 and luc , mRNA levels were normalized to the peak level on day 8 (10 A.M. time point). The y-axis scales were set at 120% maximum for all plots to allow direct comparison of mRNA extracted on days 3 and 8. The x-axis scales are given in both hours and days postfertilization to facilitate the comparison with Figure 2 . In order to show more detailed temporal profiles of mRNAs on day 3, plots with smaller y-axis scales were shown in the insets at top right corners of (A) and (C). Each plot is the average of three identical experiments, and error bars represent SEM. An identical experiment was also done at 24 °C with essentially the same results (unpublished data). Effects of Ambient Temperature on per3-luc -Mediated Bioluminescence The experiments presented above were done at 21–24 °C simply because fish survived better at these rather low temperatures (see Materials and Methods ). However, the previously documented circadian studies on zebrafish, including those involving development of rhythmicity, have been done mainly at the higher temperatures of 25–28.5 °C [ 22 , 23 , 24 , 25 ]. Since higher temperatures accelerate development in general, it was conceivable that development of bioluminescence rhythms may be faster at higher temperatures. However, it was not possible to do the same experiment at higher temperatures, because fish do not survive well in microtiter wells at temperatures higher than 25 °C. Therefore, embryos were raised in petri dishes at two different temperatures, 22 °C and 28.5 °C, while exposed to two, four, or six LD cycles. Subsequently, they were placed in microtiter wells and bioluminescence was recorded in DD at 21–24 °C ( Protocol S1 ). The majority of embryos that were raised at 22 °C were arrhythmic after they were entrained by two or four LD cycles, but they were highly rhythmic after six LD cycles ( Figure 4 A, 4 C, and 4 E; Table 2 ). This is largely consistent with the trends observed in Figure 2 and Table 1 , although more fish were rhythmic in this experiment for the two-LD and six-LD groups, and less in the four-LD group. The increased percentage of rhythmicity in this experiment for the six-LD group may be due to the fact that embryos raised in petri dishes are generally healthier than those raised in 96-well plates. Embryos raised at 28.5 °C showed significantly higher rhythmicity for the four-LD group than the same group raised at 22 °C ( p < 0.05; Figure 4 C and 4 D; Table 2 ). This result shows that embryos raised at higher temperatures can be entrained earlier than those raised at lower temperatures. Figure 4 Effects of Temperature on Development of per3-luc Expression Transgenic embryos were entrained by two, four, or six LD cycles at either 22 °C or 28.5 °C, and monitored for bioluminescence in DD at 21–24 °C. (A) Two LDs at 22 °C, (B) two LDs at 28.5 °C, (C) four LDs at 22 °C, (D) four LDs at 28.5 °C, (E) six LDs at 22 °C, and (F) six LDs at 28.5 °C. The insets in (E) and (F) show the last 3 d of the record with magnified y-axis scales. Black and white bars on top of each plot represent the times when the lights were off and on, respectively. Actual amount of bioluminescence in cps is averaged and plotted in each graph. Number of animals that were averaged is given at top right corner of each plot. Error bars represent SEM. Results of one of two identical experiments with similar results are shown here. In contrast to the effects on rhythmicity, temperature during the first several days of development seemed to have no systematic effects on periods: The two methods used for period estimation showed opposite effects of developmental temperature on periods (see the six-LD groups in Table 2 ). It should be pointed out that this may not mean that per3-luc rhythm is temperature compensated, because all the fish were monitored in the same temperature condition. Besides development of rhythmicity, the developmental profile of luc expression and baseline level of bioluminescence were also affected by prior ambient temperature. For the two-LD groups, the first and second developmental peaks came earlier in the 28.5 °C than in the 22 °C group ( Figure 4 A and 4 B). Higher temperatures also caused elevated levels of baseline bioluminescence, especially in the six-LD group ( Figure 4 E and 4 F). This higher bioluminescence cannot be caused by high specific activity of equivalent luciferase enzyme, because all of the fish were monitored at the same temperature, and the difference in bioluminescence level persisted through over 4 d of monitoring. Therefore, this difference in luminescence most likely reflects a difference in the level of luc expression. Part of this difference between the two temperature groups may be explained by the fact that animals raised at higher temperatures are more mature and therefore express more luciferase than do those raised at lower temperatures. However, 10-d-old animals raised at 22 °C, which should have reached the plateau of luminescence (see Figure 2 ), showed much lower bioluminescence than 10-d-old fish raised at 28.5 °C (compare Figure 4 E and 4 F). Therefore, maturity of animals cannot explain the difference either. It seems that animals raised at higher temperatures simply express more luciferase than do those at lower temperatures. This was confirmed by a real-time PCR experiment ( Figure 5 ). Cycling amplitudes and peak levels of per3 and luc mRNA on day 6 were much higher in fish raised at 28.5 °C than those at 22 °C. The difference was especially large for luc, for which an approximate 40-fold difference between the two temperature groups was found at Zeitgeber Time 0 (2 h after lights-on; Figure 5 C). Figure 5 Levels of per3 and luc mRNA Are Elevated by High Temperatures during Development (A) A schematic diagram showing how embryos were entrained and collected for RNA extraction. Embryos were entrained in 14:10 LD cycles (lights on at 8 A.M. ; lights off at 10 P.M. CST) at two different temperatures, 22 °C and 28.5 °C. On day 6, half of the animals were sacrificed for RNA at 10 A.M. (2 h after lights-on) and at 10 P.M. (at lights-off). The rest of the animals were transferred to DD at 22 °C at 10 P.M. on that day, and sacrificed for RNA on day 10 at 10 A.M. and 10 P.M. The white and black bars represent day and night, respectively, and the gray bars the time at which lights would have been on had the LD cycles continued. The arrowheads indicate the time at which the animals were sacrificed for RNA extraction. (B) Relative mRNA level per animal for per3 on days 6 and 10 of the experiment quantified by real-time qPCR. The levels were normalized to the value of the 10 A.M. time point on day 6 at 28.5 °C. (C) Relative RNA level per animal for luc measured from the same samples used in (B). For both (B) and (C), averages of three experiments are shown. Error bars represent SEM. Bioluminescence in the high-temperature group gradually decreased over several days after they were transferred to lower temperatures, but did not fully return to the level of the low-temperature group (see Figure 4 E and 4 F). This is consistent with per3 and luc mRNA levels determined by real-time PCR ( Figure 5 ). Taken together, high temperatures elevate the level and cycling amplitudes of per3 and luc mRNA, at least in larval fish, and this level and amplitude can gradually decrease after the animals are shifted down to lower temperatures. Optimal Condition for Recording Larval per3-luc Rhythms Fold amplitudes of bioluminescence rhythms were higher when embryos were entrained by LD cycles for 6 d at 22 °C rather than at 28.5 °C (see Figure 4 E and 4 F). Furthermore, survival of the animals was better if they were raised at 22 °C than at 28.5 °C (96.9% and 0% survival, respectively, on day 7). Therefore, embryos were entrained by six LDs at 22 °C in a petri dish, and tested for bioluminescence rhythms in DD ( Protocol S1 ). In this experiment, embryos survived better than they did when they were placed in 96-well plates from day 1 onward (89.6% survival on day 12 in this experiment, compared to 71.5% on day 10 for the experiments presented in Figure 2 ). Furthermore, 88.1% ( n = 42) of the fish were rhythmic with a 25.2 ± 0.7 h (mean ± standard deviation) period under this condition, and their rhythms persisted for 6 d, albeit with some damping ( Figure 6 A and 6 B). In addition, per3-luc rhythms were tested in LD ( Protocol S1 ). Amplitudes of bioluminescence rhythms in LD were higher than in DD, although they also damped slightly, possibly due to substrate deprivation ( Figure 6 C and 6 D). A slightly higher percentage of fish was rhythmic in LD (95.0%, n = 179) compared to DD, although the difference was not significant ( p > 0.1, G-test). The waveform of the rhythm in LD was different from that in DD: The ascending part of the wave that happens during the day was steeper in LD than in DD ( Figure 6 ), suggesting that light may induce transcription of per3 . Figure 6 Bioluminescence Rhythms Mediated by per3 - luc in DD and LD Measured for Six Days (A) Average plot of bioluminescence rhythms in DD. Animals were entrained in 14:10 LD cycles for 6 d at 22 °C and tested for approximately 6.5 d in DD. (B) Representative plot of bioluminescence rhythm in DD for an individual. (C) Average plot of bioluminescence rhythms in 14:10 LD cycles. Animals were entrained in 6 LD cycles at 22 °C prior to the monitoring. (D) Individual plot of bioluminescence rhythm in LD cycles. For each of DD and LD experiments, two experiments have been performed with essentially the same results. Only one of two experiments is shown for each of DD and LD. The first 12 h of data were deleted from each plot. Black and white bars on top of each plot represent the time when lights were off and on, respectively. Numbers of animals that were averaged is given at top right corner of (A) and (C). Error bars represent SEM in (A) and (C). Discussion The per3-luc transgenic zebrafish system presented here is unique, because it is the only vertebrate system in which circadian gene expression in the whole animal can be studied in a high-throughput manner. This property of per3-luc in combination with zebrafish genetics makes the transgenic fish suitable for mutagenesis screening for circadian mutants. Embryos were tested initially, because it is more efficient to screen embryos than older animals, and movements of older animals can cause noise in bioluminescence signals [ 28 ]. However, the current study clearly demonstrates that monitoring embryos is not an option for any circadian studies using these lines. Of the conditions tested, raising fish to 6 d of age at 22 °C under LD cycles produced the most robust free-running rhythms. Rhythms measured under LD cycles were even stronger, and this condition has been used successfully in previous screens [ 15 ]. In addition to mutagenesis screening, it will facilitate other studies such as circadian organization of central and peripheral oscillators, entrainment pathways by various environmental cues, and physiological effects on circadian rhythms as have been studied in other organisms [ 17 , 18 , 29 , 30 , 31 , 32 , 33 ]. Development of per3 RNA Cycling in Larval Zebrafish Larval fish that experienced an LD-DD transition later during development were more rhythmic than those shifted to DD earlier. This may mean that there is a critical developmental period after which per3-luc rhythms can be entrained. Alternatively, per3-luc rhythms might damp so fast that rhythms entrained a few days earlier could not be detected. We think the latter possibility unlikely for the following reasons: per3-luc -mediated bioluminescence rhythms persisted reasonably well for at least 6 d in DD if entrained properly ( Figure 6 A); at 28.5 °C, the rhythmicity of larval fish entrained for 4 d was comparable to that of fish entrained for 6 d, suggesting that bioluminescence rhythms do not damp in 2 d ( Table 2 ); and luc mRNA cycling was almost undetectable during the first few days of development (see Figure 3 C), and this low-amplitude fluctuation cannot give rise to high-amplitude oscillations unless there is a separate rhythm-amplifying mechanism operating during development, such as synchronization of cellular oscillators. The results presented here, namely the gradual development of rhythmicity and responsiveness to entraining stimuli during the first several days of development and requirement of light signal, is consistent with previous observations on rhythms of melatonin production, locomotor activity, and the cell cycle in larval zebrafish [ 23 , 24 , 25 ]. It is also consistent with several studies in other vertebrate species involving gene expression [ 34 , 35 , 36 ] and physiological rhythms [ 37 ]. In both insects and mammals, free-running behavioral rhythms can develop normally in the absence of entraining signals, but their phases are not synchronized [ 38 , 39 , 40 ]. DeLaunay et al. [ 22 ] reported that per3 RNA detected by in situ hybridization cycled synchronously from day 1 of development [ 22 ]. In our hands, however, per3 and per3 -driven luc RNA in 3-d-old larval fish detected by real-time PCR were not significantly rhythmic. Low-amplitude oscillations of both mRNAs may exist at this early developmental stage, because a similar trend (high in the morning) was observed in all three independent experiments done. However, this does not mean that the low-amplitude per3 RNA oscillations that occur earlier during development can amplify into high-amplitude ones without any entrainment by LD cycles. The increase of cycling amplitude over the course of several days of development in LD may happen within each cell that expresses per3 . Alternatively, cycling amplitude may increase because various cellular oscillations present become synchronized. However, it was not only the cycling amplitude that increased with age, but the overall levels of per3 mRNA also. Therefore, more cells may start expressing per3 with high-amplitude oscillations later during development. This rather simple scenario is in fact what happens in developing Drosophila; as soon as the central pacemaker Lateral Neurons start expressing the PERIOD protein, the molecular rhythm is entrainable, and so is the eventual behavioral rhythmicity [ 40 , 41 ]. In any case, increasing amplitudes within cells, synchronization of oscillators, and more cells with high-amplitude oscillations are not mutually exclusive. It is worth mentioning that rhythms of melatonin release starts as early as day 2 postfertilization, and this roughly corresponds to the time when the light-sensitive pineal gland is formed [ 23 ]. Another photosensitive organ, the retina, becomes photoresponsive as early as day 3 postfertilization [ 42 ]. Therefore, these organs become photosensitive, and/or develop rhythmicity prior to robust oscillations of per3 RNA. Again, clocks in some tissues may develop earlier than in other tissues. It is also possible that per3 may not be expressed in all the clock cells. The biological significance of the developmental profile of per3 -driven luc expression is not known. Most animals seemed to hatch between the first minor developmental peak and the subsequent trough (unpublished data). However, hatching itself is unlikely to induce per3 expression in early embryos, because dechorionated per3-luc embryos showed developmental profiles of bioluminescence similar to those of nondechorionated siblings, albeit with an accelerated second rise of bioluminescence (unpublished data). Larval fish after the hatching period are supposed to have completed most of their morphogenesis and start swimming actively [ 43 ]. It may be that per3 is important for rhythmic processes specific to hatched animals, such as behavioral rhythms [ 24 ]. Consistency between per3 Expression and Bioluminescence The developmental and circadian profiles of per3 -driven bioluminescence largely reflected endogenous per3 expression. However, the amplitude of bioluminescence cycling was greatly reduced compared to that of per3 or luc RNA. In general, proteins synthesized from cycling mRNAs show amplitude reduction and phase delay due to protein stability [ 44 ]. Consistently, luciferase-reporting studies in other organisms also showed dampening of cycling. For instance, in the per - luc transformants of Drosophila, approximately 6-fold cycling of luc RNA was reduced to 3- to 4-fold bioluminescence rhythms [ 27 ]. Similar reduction in cycling amplitude was observed for suprachiasmatic nuclei from the per1-luc mouse [ 18 ]. However, the reduction of amplitude was even more dramatic in per3-luc larval fish. This may mean that the luciferase protein is somehow more stable in larval fish than in flies or mice. It should be noted that the luciferase protein itself is quite stable, but the enzymatic activity of this protein is unstable [ 28 , 45 ]. Therefore, the apparent stability of the luciferase protein in larval fish may be in fact stability of luciferase activity. Besides the difference between bioluminescence and mRNA, there were differences between per3 and luc mRNA, measured by real-time qPCR. The difference between days 3 and 8 of development was much larger for luc than per3 . Furthermore, the difference between the 22 °C and 28.5 °C groups was larger for luc than for per3 . These differences may be due to positional effect of the insert in the particular transgenic line used in this study. However, differences in bioluminescence levels between the first few days and older fish were found in two other independent per3-luc lines (unpublished data). Therefore, it is more likely that these differences between the two mRNA species reflect a property of the transgene itself. Although the BAC transgene used in this study has approximately 26 kb of upstream sequences, there may be critical sequences missing from this transgene, such as the first coding exon. Alternatively, posttranscriptional modification could be responsible for the difference. Posttranscriptional control of clock gene mRNA expression has been documented elsewhere [ 18 , 28 , 46 , 47 ]. Effects of Ambient Temperature on per3 Expression Increasing ambient temperature had three effects on development of per3 expression: It increased peak levels and cycling amplitudes of mRNA, and accelerated the developmental profile and development of responsiveness to the entraining stimuli. The latter two effects presumably are due simply to the fact that development is accelerated by higher temperatures. These results are not consistent with the report that development of a cell-cycle rhythm in larval fish was not affected by ambient temperature [ 25 ]. The increase in peak level and cycling amplitudes of per3 mRNA induced by increased temperature is mostly independent of developmental speed. The peak per3 mRNA level in the high-temperature group on day 6 was higher than that in the low-temperature group on day 10, when bioluminescence levels should have reached the plateau (see Figure 5 ). Ambient temperatures are known to affect levels of clock gene expression in Drosophila and Neurospora [ 47 , 48 , 49 ], and this could be the mechanism for clock resetting by temperature shifts [ 49 ]. Since our study was done on developing animals, it remains to be seen whether the increase in per3 expression by elevated temperatures holds true in adults. Furthermore, it was not possible to test larval fish at 28.5 °C, and therefore it is not known whether the per3 mRNA rhythm is temperature-compensated. Levels of per3 and luc mRNA in fish raised at higher temperatures did not fully return to the levels in fish raised at lower temperatures, even after 4 d at lower temperatures. Therefore, there may be a mechanism for maintaining circadian periods despite change in per3 expression levels. Materials and Methods Animals Animals used in this study were derived from the University of Oregon AB strain. Adults were kept under 14:10 LD (lights on at 8 A.M. ; lights off at 10 P.M. CST) cycle, and group-housed in plastic tanks in a Z-MOD holding system (Marine Biotech) with recirculating filtered water at about 28.5 °C. They were fed commercial flake food in the morning, baby brine shrimp at midday, and adult brine shrimp in the evening. Experimental protocols were approved by the Institutional Animal Care and Use Committee. Embryos were collected from naturally breeding fish in the morning, by plastic mesh traps that prevented parents from eating their progeny [ 23 ]. For microinjection, one- to two-cell-stage embryos were required. Therefore, male and female breeders were separated by a divider when they were placed in a trap. By removing the divider, fish were allowed to breed just before microinjection was performed [ 50 ]. Construction of per3-luc transgene First, a zebrafish BAC library was screened for BACs containing 5′ coding sequences of per3 [ 22 ] using a PCR-based screening kit (Incyte Genomics, Wilmington, Delaware, United States). One of two such clones, 8M06, was used for the construction of the transgene. The primers used for the screening were: forward, 5′- CCAGTAAAACGTCGTCGTCA-3′; reverse, 5′- GTCTGGGCCTGGAGAAGAGT-3′. The per3 sequence from the initiation codon to the end of the first coding exon was replaced by a gene cassette containing the luc and Km r genes by homologous recombination in E. coli (see Figure 1 ) [ 51 , 52 ]. The luc / Km r gene cassette was constructed as follows. First, the Km r gene was cloned into NotI and SacI sites of pBluescriptSK+ (Stratagene, La Jolla, California, United States); then a HindIII-BamHI fragment containing luc and the SV40 polyA signal from pGL3-Basic (Promega, Madison, Wisconsin, United States) was cloned upstream of Km r into HindIII and BamHI sites of pBluescriptSK+; finally, a NotI site between Km r and the polyA signal was destroyed by digesting the clone with NotI and BamHI, followed by treatment with the Klenow fragment, and ligation of these blunted ends together. Using this gene cassette as a template, a PCR fragment flanked by approximately 50-bp homology arms was amplified. The primer sequences used for the PCR reaction were: forward, 5′- GGGTTGTGAATCAGATCTTCAGTAGAGGAGGACAGGAGATCTCACAGGGAATGGAAGACGCCAAAAACATAAAGAAAG-3′; reverse, 5′- GTGCAGATTAAGTCAAATTCCACATAAAAAAAGCCACATTTCAAGTGTAC CGTTAATAATTCAGAAGAACTCGTC-3′. The forward primer contains 25 bp of sequence from the 5′ end of the luc coding sequences flanked by a 53-bp overhang corresponding to sequences just upstream of the initiation codon of per3 . The reverse primer consists of a 50-bp overhang that corresponds to the intron sequences just downstream of the first coding exon, and 25 bp of sequence from the 3′ end of Km r as the primer. The PCR fragment was purified and electroporated into DH10B cells containing the BAC and the plasmid pBAD-αβγ [ 51 ] as a source of recombinase genes. Cells in which the BAC was successfully modified were selected by kanamycin. The luc sequences in the modified BAC clones were checked for PCR errors by sequencing. One PCR error that resulted in a Val 217 to Ala change in the luciferase protein sequence was found. However, this is a conservative change, and fish injected with this construct showed bioluminescence above background. Therefore, it was judged that luciferase encoded by this construct can still function. Generation of per3-luc transgenic lines The per3-luc transgene was purified, linearized by NotI digestion, and injected into one- to two-cell-stage zebrafish embryos according to [ 53 ] with minor modifications. Injected embryos were raised to adulthood and individually bred to a wild-type fish or pairwise bred to each other. The progeny were tested for the presence of the transgene by PCR. PCR primers used for the screening were: Per35F, 5′- GCACCAGTAAAACGTCGTCA-3′; Per33R, 5′- TCATTCTCACTGGCAGAGCA-3′; Luc5R2, 5′- GTTTTAGAATCCATGATAATA-3′; Kan3F2, 5′- CTTTTTGTCAGAAGACCGACC-3′. The approximate positions of these primers with respect to the construct are shown in Figure 1 . Transgenic fish were identified as those fish that gave an approximately 600-bp PCR product with the Per35F/Luc5R2 primer combination, and an approximately 800-bp product by the Kan3F2/Per33R combination, when their genomic DNA was used as templates. Nontransgenic fish gave no products with either of these primer combinations. In vivo measurement of bioluminescence rhythms Embryos or larval fish were placed individually in every other well of a white 96-well Optiplate (Perkin-Elmer, Wellesley, California, United States) with 200 μl of Holtfreter solution (7.0 g of NaCl, 0.4 g of sodium bicarbonate, 0.2 g of CaCl 2 , and 0.1 g of KCl [pH 7.0] in 2 l of ddH 2 O) aerated overnight and containing 0.5 mM D-luciferin potassium salt (Biosynth, Naperville, Illinois, United States) and 0.013% Amquel Instant Water Detoxifier (Kordon brand; Novalek, Hayward, California, United States). Once loaded with animals, four such plates were subjected to automatic monitoring of bioluminescence every 30 min by the Topcount multiplate scintillation counter (Perkin-Elmer) equipped with six detectors and plate stackers. The room temperature was set at 21–22 °C, and the machine at 24 °C. However, due to the heat created by the machine, temperature at the bottom of the stacker was 1–2 °C higher than the room temperature. In order to minimize high background counts under lighted conditions, each plate was dark-adapted for approximately 5 min before being counted for bioluminescence. Each well was counted for 4.8 s every 30 min. The plates were illuminated with two white fluorescent lamps, each facing the left or right side of the stacker. The approximate intensity of the light that reached the plates was 17–35 lux, depending on the position of the plates within the stacker. Experimental protocol The experimental protocol for each experiment involving in vivo bioluminescence measurement is described in Protocol S1 . Bioluminescence data analyses Bioluminescence data from the Topcount were imported into Microsoft Excel 2000 by the Import and Analysis macro (kindly supplied by Steve Kay, Scripps Institute). In many of the experiments performed on the Topcount, some plates were placed in the machine several days earlier than others in order to monitor fish that experienced different numbers of LD cycles ( Protocol S1 ). Therefore, at the end of the recording period that typically lasted for approximately 2 wk, many fish that were placed in the machine earlier had been dead for a few days. Since only the data up to 10-d old larval fish were analyzed for the experiments presented in Figures 2 and 4 , simple observation of fish after the recording period could overestimate the number of fish that had died while the 10-d worth of data were collected. When a per3-luc fish dies in luciferin solution, it emits a burst of high bioluminescence counts (> 2,000 cps). This burst of luminescence is typically followed by a low background level of luminescence (< 50 cps). Furthermore, intermediate levels of spikes were also found in many plots just before the burst of high bioluminescence. Therefore, in order to eliminate data from dead fish, data that exceeded 5,000 cps, or those that went down below 50 cps, at any point of the analyzed portion of the data were first discarded. Then an averaged plot of the remaining data from each clutch of embryos was examined, and the highest count on days 8–9 was determined for experiments presented in Figure 2 . For the other experiments, highest counts from the entire averaged plots except for day 1 of the record were determined. Then data that exceeded twice that value were also discarded as those with medium-sized spikes. This second round of data elimination was done in this way because overall luminescence counts varied among different clutches, possibly due to varying sizes of eggs laid. It should be noted that this procedure also eliminated records from fish that were alive, but that showed one or more transient spikes of bioluminescence. Those data with transient spikes were eliminated anyway, because the spikes can severely affect the accuracy of the data analysis program. Period and rhythmicity for each animal were determined by a macro [ 54 ] based on MatLab 6.5 (Mathworks, Natick, Massachusetts, United States). With this macro, periods were determined by the maximum entropy spectral analysis (MESA) and autocorrelation, and rhythmicity by autocorrelation. Some fish that were apparently arrhythmic by visual examination of the plot gave rhythm indexes with a confidence interval higher than 95% by autocorrelation due to spurious peaks and small confidence intervals. Autocorrelation plots of these fish, however, were almost always nonsinusoidal and/or did not have five clear peaks (one at the center and two on each side of the center peak). Therefore, each set of data was judged blindly by three people as to whether its autocorrelation plot was sinusoidal with five peaks. Fish were judged as rhythmic only if two to three people found their autocorrelation plots sinusoidal, and their rhythm statistic values exceeded 1. Real-time qPCR Total RNA was extracted from 9–42 embryos or larval fish raised in petri dishes using TRIzol reagent (Invitrogen, Carlsbad, California, United States). The number of animals used for each extraction was recorded. Once extracted, total nucleic acid concentration was determined by a spectrophotometer. In order to prevent genomic DNA contamination, RNA samples were treated with Turbo DNA-free (Ambion, Austin, Texas, United States), and the concentration determined again by a spectrophotometer. Total RNA (0.5–1 μg) was subjected to cDNA synthesis by Superscript II Reverse Transcriptase (Invitrogen) using Oligo (dT) 12–18 (Invitrogen) as the primer in a 25–40 μl reaction volume. Real-time PCR was performed in a 25 μl reaction volume containing a probe, forward and reverse primers, and qPCR Mastermix according to the manufacturer's instruction (Eurogentec, Seraing, Belgium). Each reaction was quadrupled in order to minimize pipetting errors. The primers and TaqMan MGB probes for per3 and luc were designed and synthesized by the Assays-by-Design Gene Expression service (Applied Biosystems, Foster City, California, United States): per3 forward, 5′- GCCCTGGCAGCACCA-3′; per3 reverse 5′- GAAAGCTGGAGGACGAGGAA-3′; probe, 5′-6-FAM- CTAAGAGCTCAAAATCC-NFQ-3′; luc forward, 5′- GCAGGTGTCGCAGGTCTT-3′; luc reverse, 5′- GCGACGTAATCCACGATCTCTTTT-3′; probe, 5′-6-FAM- TCACCGGCGTCATCG-NFQ-3′. The ABI Sequence Detection System 7000 (Applied Biosystems) was programmed to perform the following protocol: 50 °C for 2 min, 95 °C for 10 min, followed by 40 cycles of 95 °C for 15 s and 60 °C for 1 min. In this study, relative amount of per3 or luc cDNA per animal was calculated by the standard-curve method [ 55 ] rather than by normalizing those RNA species to a constitutive control gene, for the following reasons: Both per3 and luc were compared between two different developmental stages as well as among different times of the day. It was also important to calculate the amount of each mRNA species per animal in order to compare these data to bioluminescence data. The amount of a specific control RNA, as well as the total RNA, may differ among fish of different ages, in which case RNA per animal cannot be calculated by the relative quantification method using a constitutive control. As a concentration standard, a single-stranded DNA oligonucleotide of known concentration was used for each gene. These oligonucleotides span from the 5′ end of the forward primer to the 5′ end of the reverse primer, and including 75 bp for per3 and 110 bp for luc (Biosource, Camarillo, California, United States). The standard concentration was varied from 10 2 to 10 7 copies per reaction in 10-fold increments. For every qPCR experiment, reactions for standards were performed in four replicates along with reactions for cDNA samples. Statistics To test whether percentages of rhythmic fish among different experimental groups were equal, the G-test was performed using Microsoft Excel 2000 according to Sokal and Rohlf [ 56 ]. If multiple tests were performed for a set of data, critical value of the G-statistic was adjusted for the experimentwise error rate [ 56 ]. For all the other numerical data, JMP 3.1.5 (SAS Institute, Cary, North Carolina, United States) was used for the following tests: Each set of data were first subjected to the test for normality. If the data were normally distributed, the one-way analysis of variance or the t -test was performed. The nonparametric Wilcoxon/Kruskal-Wallis test was performed on data that were not normally distributed even after various transformations (logarithmic, square root, and inverse) were tried. Where multiple tests were performed on a set of data, the experimentwise error rate (α) was adjusted by the Dunn-Sˇidák method [ 56 ]. Supporting Information Protocol S1 Experimental Protocol for Bioluminescence Experiments (27 KB DOC). Click here for additional data file. Accession Numbers The GenBank ( http://www.ncbi.nlm.nih.gov/ ) accession numbers of the sequences discussed in this paper are per 3 cDNA (NM_131584) and BAC clone CH211–138E4 (AL929204). The Ensembl ( http://www.ensembl.org/Danio_rerio/ ) ID of the flanking gene of per 3 mentioned in Figure 1 is ENSDARG00000023492.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC546037.xml
548146	Induction of chondro-, osteo- and adipogenesis in embryonic stem cells by bone morphogenetic protein-2: Effect of cofactors on differentiating lineages	Background Recently, tissue engineering has merged with stem cell technology with interest to develop new sources of transplantable material for injury or disease treatment. Eminently interesting, are bone and joint injuries/disorders because of the low self-regenerating capacity of the matrix secreting cells, particularly chondrocytes. ES cells have the unlimited capacity to self-renew and maintain their pluripotency in culture. Upon induction of various signals they will then differentiate into distinctive cell types such as neurons, cardiomyocytes and osteoblasts. Results We present here that BMP-2 can drive ES cells to the cartilage, osteoblast or adipogenic fate depending on supplementary co-factors. TGFβ 1 , insulin and ascorbic acid were identified as signals that together with BMP-2 induce a chondrocytic phenotype that is characterized by increased expression of cartilage marker genes in a timely co-ordinated fashion. Expression of collagen type IIB and aggrecan, indicative of a fully mature state, continuously ascend until reaching a peak at day 32 of culture to approximately 80-fold over control values. Sox9 and scleraxis, cartilage specific transcription factors, are highly expressed at very early stages and show decreased expression over the time course of EB differentiation. Some smaller proteoglycans, such as decorin and biglycan, are expressed at earlier stages. Overall, proteoglycan biosynthesis is up-regulated 7-fold in response to the supplements added. BMP-2 induced chondrocytes undergo hypertrophy and begin to alter their expression profile towards osteoblasts. Supplying mineralization factors such as β-glycerophosphate and vitamin D 3 with the culture medium can facilitate this process. Moreover, gene expression studies show that adipocytes can also differentiate from BMP-2 treated ES cells. Conclusions Ultimately, we have found that ES cells can be successfully triggered to differentiate into chondrocyte-like cells, which can further alter their fate to become hypertrophic, and adipocytes. Compared with previous reports using a brief BMP-2 supplementation early in differentiation, prolonged exposure increased chondrogenic output, while supplementation with insulin and ascorbic acid prevented dedifferentiation. These results provide a foundation for the use of ES cells as a potential therapy in joint injury and disease.	Background Articular cartilage is composed of extracellular matrix (ECM), the matrix-secreting chondrocyte and water, which all account for the tissue's characteristic rigidity as well as its flexibility. These features are necessary in order to warrant life-long survival of the cartilage tissue, especially in the joint where it has to endure pressure forces caused by movement. Chondrocytes arise from a mesenchymal progenitor during development, the same progenitor that gives rise to other mesenchymal cell types including osteoblasts, adipocytes and myocytes. Bone formation can either be endochondral, when chondrocytes mature and calcify to provide a matrix for the invading osteoprogenitors, or intramembraneous involving ossification directly from a mesenchymal ancestor. All these diverse cell types may arise from the same precursor, but are distinguished by specific morphological features and with that, a certain set of characteristic proteins including transcription factors that control their differentiation. Two chondrocyte-specific transcription factors have been identified, Sox9, a member of the SOX-family of transcription factors, and Scleraxis, a member of the basic helix-loop-helix transcription factors [ 1 , 2 ]. However, most of the exclusive markers for cartilage tissue reside in the ECM. The predominant form of collagen in mature cartilage is collagen type IIB, whereas the alternatively spliced collagen IIA is found primarily during development [ 3 ]. Aggrecan is the major proteoglycan species in cartilage [ 4 ]. The transition of chondrocytes into hypertrophy is distinguished by a change in expression of Cbfa1, the osteoblast-specific transcription factor, which is also switched on during intramembraneous ossification. Distinctive to cartilage, the major collagen molecule in osseus matrix is collagen type I. In contrast, the transcription factors controlling adipogenesis are C/EBPα and PPARγ, which transactivate subsets of genes as a function of either trans-acting factor alone or requiring the co-operative effort of both [ 5 ]. C/EBPα is known to bind to and transactivate particularly the promotors of the SCD1, aP2 and the Glut4 genes [ 6 , 7 ], all highly characteristic of the adipocyte phenotype. For decades, the treatment of degenerative cartilage and bone diseases has been a challenge for orthopaedic surgeons due to the apparent inability of cartilage and bone to repair itself. Arthritis, a degenerative joint condition, is one of the most prevalent chronic health conditions in North America. Arthritis can devastate people, but to date there is no effective therapy available and patients can only be helped by surgical joint replacement. An inherent major concern is the limited availability of autografts, which significantly reduces the choice of treatable defects. However, new approaches to cell grafting are being developed in this field: increased yields of cells are achieved by the usage of bioreactors and growth factor administration, such as TGFβ 1 and BMPs [ 8 , 9 ]. Additionally, stem cells are being discovered as a new source of transplantable material. Embryonic stem cells represent a valuable source for cell transplantation since their characteristic features include an unlimited self-renewing capacity and a multilineage differentiation potential [ 10 , 11 ]. In fact, ES-derived glial precursors and cardiomyocytes have been successfully transplanted, integrated and shown to be functionally active in the transplantation site [ 12 , 13 ]. The yield of differentiation of ES cells into an intended lineage can be greatly enhanced by the addition of growth factors or induction substances. Whereas protocols for the differentiation of cardiomyocytes, neuronal cell types, insulin-producing cells or adipocytes from ES cells have been available for many years [ 14 - 17 ], only recently their differentiation into elements of the skeleton has been reported [ 18 - 20 ]. Our group has previously shown that vitamin D 3 forces ES cells to undergo osteogenesis [ 18 ]. Kramer et al. have reported in 2000 that BMP-2 pushes ES cells to the chondrogenic fate when added during days 3–5 of EB differentiation [ 21 ]. Those ES-derived chondrocytes possess a certain plasticity to undergo hypertrophy and calcify [ 22 ]. We show here, that prolonged treatment of differentiating ES cell cultures with BMP-2 in synergy with TGFβ 1 , insulin and ascorbic acid leads to improved chondrogenesis in vitro . Compared to the brief supplementation described by Kramer et al. [ 21 ], the expression of chondrocyte-specific marker genes was highly up-regulated while proteoglycan content revealed an increased chondrocytic yield from 7.26% to 57.03%. As described by other groups [ 22 ], ES-derived chondrocytes become hypertrophic and calcify. However, spontaneous calcification did not reach mineralization levels that are found in vitamin D 3 induced ES-derived osteoblasts [ 18 ]. Yet, supplementation of chondrocyte-cultures with β-glycerophosphate, ascorbic acid and vitamin D 3 starting at day 20 rescued the osteoblast phenotype. In many differentiations, we also observed an accumulation of lipid droplets and an up-regulation of adipocyte-specific genes. This direction towards adipocyte differentiation varied with the use of specific co-factors, suggesting that in the future, such spurious differentiation may be controlled, once the pathways involved in adipogenesis are better understood. Results Characterization of chondrocyte-like cells derived from ES cells Embryonic stem cell cultures supplemented with BMP-2, TGFβ 1 , insulin and ascorbic acid show typical morphological changes compared to the untreated cultures. Starting with the fourth week of culture, aggregates consisting of small round cells formed in the supplemented cultures, which stained positive with alcian blue (fig. 1A ). Little alcian blue staining was seen in control cultures (fig. 1B ). Polygonal cells, which could also be found in treated cultures, did not stain with alcian blue. Significant immunostaining for the collagen type II (COL II) protein was observed at day 32 in treated cultures corresponding to the active secretion and formation of an extracellular matrix found with chondrocytes. The COL II antibody identified the fibrillary organization of the collagen molecules in the extracellular matrix (fig. 1C ). Chondrogenic differentiation was confirmed by positive immunostaining for adult proteoglycans (fig. 1D ), which is detectable in the aggregates identified by alcian blue staining. The distribution was associated with the extracellular matrix similar to that found with the COL II antibody. Staining appeared to be diffuse as extracellular matrix and not individual cells are stained. Figure 1 Morphology and characteristics of ES-derived chondrocytes after 32 days of culture. (A, B) Determination of proteoglycans in EBs with alcian blue in chondrocyte cultures induced with TGFβ 1 [2 ng/ml] and BMP-2 [10 ng/ml] from d3–5 of culture and with BMP-2 [10 ng/ml], ascorbic acid [50 μg/ml] and insulin [1 μg/ml] from day 5 onwards (A) compared to control cultures (B). Bar = 8.3 μm. (C, D) Analysis of cartilage-specific matrix proteins in 32 day old EBs induced with the same supplements as in (A, B) by means of immunohistochemistry. Staining with anti-collagen type II (C) and anti-cartilage proteoglycan (D), respectively, both visualized by a secondary AlexaFluor 488 conjugated antibody. Bar = 106 μm. (E) Concentration-dependent effect of BMP-2 on chondrocyte-specific gene expression in 32-day old EBs. Values of cultures supplemented with 10 ng/ml BMP-2 are shown compared to 2 ng/ml as used by Kramer et al. [2000], which was set as 1. Values represent means of three independent experiments ± standard deviation, obtained by quantitative RT-PCR analysis. P < 0.01; P < 0.001. (F) Proteoglycan content of EB extracts on day 32. BMP-2 directs increased proteoglycan synthesis in both 2 ng/ml and 10 ng/ml. P < 0.1; *P < 0.01. Mean ± standard deviation, n = 3. Dependence of gene expression patterns and proteoglycan synthesis on BMP-2 Quantitative RT-PCR was used to examine the variability of RNA expression of various cartilage-specific genes in response to BMP-2. Previously, chondrogenesis was induced in ES cells using 2 ng/ml BMP-2 on days three to five of EB formation, when early mesodermal markers such as Brachyury and BMP-4 are expressed [ 21 , 23 , 24 ]. To improve chondrogenesis, we investigated whether a higher concentration of BMP-2 (10 ng/ml) could increase chondrocyte-specific gene expression. Total RNA was extracted on day 32 and quantitative real-time PCR was carried out. Expression of genes of interest in cultures supplemented with 10 ng/ml BMP-2 was normalized to GAPDH expression and compared to cultures supplemented with 2 ng/ml (fig. 1E ), which were set as 1. Expression of the small leucine-rich proteoglycans biglycan and decorin was increased 1.6 fold upon supplementation with 10 ng/ml BMP-2 (P < 0.01). Neither link protein expression nor Sox9 or scleraxis expression were affected by the higher dosage. However, expression of aggrecan and the collagen type II isoforms A and B, which are specific for mature chondrocytes, were significantly increased compared to the low BMP-2 concentration described by Kramer et al. (P < 0.001). The degree of chondrogenic differentiation under influence of BMP-2 was further quantified by metachromatic detection of proteoglycans (fig. 1F ). Secreted proteoglycans were extracted on day 32 of culture with guanidine/HCl and combined with dimethylmethylenblue. Based on our measurement of aggrecan, the synthesis of proteoglycan proteins is also increased under the influence of BMP-2. BMP-2 at 2 ng/ml initiated an induction in proteoglycan synthesis of about 2 fold (P < 0.01), whereas 10 ng/ml BMP-2 increased the proteoglycan content of EBs 2.3 fold (P < 0.1) compared to non-treated controls. This data shows, that BMP-2 causes the induced synthesis of negatively charged extracellular matrix, characterized by proteoglycans. Additive effect of TGFβ 1 , insulin and ascorbic acid on BMP-2 induced chondrogenesis Since BMP-2 is believed to play a role in late chondrogenesis [ 25 ], we studied the effect of prolonged BMP-2 supplementation beyond day 5 of culture. Additionally, the anabolic effect of insulin and ascorbic acid and the influence of the growth factor TGFβ 1 on BMP-2 induced differentiation was determined quantitatively by PCR analysis at various stages throughout EB differentiation. Table 1 shows the particular combinations of medium supplements used at different culture stages of the 'hanging drop' protocol used for differentiation, in which day 1–3 represent the hanging drop stage. During days 3–5 the resulting embryoid bodies are cultured in suspension and then plated onto tissue culture treated plastic ware on day 5. Applied concentrations were 10 ng/ml BMP-2, 2 ng/ml TGFβ 1 , 1 μg/ml insulin and 50 μg/ml ascorbic acid. Dexamethasone, which is also known to be a chondro-inducing agent [ 26 ], did not evoke a mentionable chondrogenic response in the D3 ES cell line (data not shown). Figure 2 shows changes in cartilage-specific gene expression under the influence of various differentiation co-factors throughout the 32 days of culture. TGFβ 1 (combination B) evoked a 2-fold increase in collagen II expression for both splice forms. The synergistic effect of both growth factors, TGFβ 1 and BMP-2 (combination C) began to show an increased expressions of aggrecan, link protein and COL IIA compared to cultures that were treated with one supplement alone. Additional supplementation with insulin and ascorbic acid between culture days 3 and 5 (combination D) barely increased aggrecan, link protein and COL II expression, but when given from day 5 onwards enhanced both the BMP-2 and TGFβ 1 effects (supplement combinations E and H). Addition of insulin and ascorbic acid to BMP-2 induced cultures increased collagen type IIA and aggrecan expression minimally to 1.2-fold, link protein and collagen type IIB were induced 2.7- to 2.8-fold compared to BMP-2 alone. The TGFβ 1 response was increased 3- to 4-fold. When cultures were induced with BMP-2 and TGFβ 1 in suspension (d3–5) and supplemented with insulin and ascorbic acid starting on day 5 onwards (combination F), aggrecan, link protein and COL II were up-regulated 10-fold compared to controls. Table 1 Combinations of medium supplements used at different culture stages Supplement combination Day 3–5 (hanging drops) Day 5 onwards (attached culture) A BMP-2 B TGFβ 1 C BMP-2 TGFβ 1 D BMP-2 TGFβ 1 insulin ascorbic acid E BMP-2 insulin ascorbic acid F BMP-2 TGFβ 1 insulin ascorbic acid G BMP-2 TGFβ 1 insulin ascorbic acid BMP-2 H TGFβ 1 insulin ascorbic acid Applied concentrations were 10 ng/ml BMP-2, 2 ng/v ml TGFβ 1 , 1 μg/ml insulin and 50 μg/ml ascorbic acid. The end of the culture period varied with every experiment carried out. Figure 2 Changes on cartilage markers in response to various combinations of BMP-2, TGFβ 1 , insulin and ascorbic acid as outlined in table 1. (A) Diagram of expression niveaus of cartilage-specific genes of 32 day old EBs obtained by quantitative RT-PCR and standardized to GAPDH. Untreated control values were set as 1. Mean ± standard deviation, n = 3. P < 0.1; P < 0.01; P = 0.001. (B) Proteoglycan content of extracts of 32 day old EBs treated with different combinations of BMP-2, TGFβ 1 , insulin and ascorbic acid (see table 1). Ctrl = control. Means ± standard deviation, n = 3. P < 0.1; P < 0.01; P = 0.001. n.d. = not determined. Surprisingly, when BMP-2 supplementation was maintained throughout the culture period (combination G), a dramatic increase of marker gene expression was seen. Aggrecan and COL IIB were up-regulated over 80-fold above controls and link protein and COL IIA expression was increased to 35- and 16-fold, respectively. Here, scleraxis and Sox9 expression behaved conversely. Supplement combination G showed significant decreases of scleraxis and Sox9 expression to 50 and 87% of the control, respectively (P = 0.001). Deferral of the expression of these transcription factors in favour of chondrogenic differentiation characterizes BMP-2 induced differentiation, whereupon chondrogenic processes are furthermore enhanced by TGFβ 1 , insulin and ascorbic acid. The determination of proteoglycan content was used to confirm the results obtained by quantitative RT-PCR for all supplement combinations (fig. 2B ). TGFβ 1 alone (combination B) did not alter proteoglycan levels compared to controls, and in combination with BMP-2 decreased the proteoglycan content of EBs compared to BMP-2 alone (combination C). Insulin and ascorbic acid did not enhance the proteoglycan synthesis in combination with TGFβ 1 , but caused a 4.2-fold increase in combination with BMP-2 (combination E, P = 0.01). In agreement with aggrecan gene expression, combinations F and G generated a massive 7-fold induction of proteoglycans in EBs (P = 0.001). Genetically manipulated ES cells that express GFP under the control of chondrocyte-specific aggrecan promotor were then used to quantify chondrocyte yield using fluorescence-activated cell sorting. ES cells were differentiated along the chondrocytic lineage using BMP-2 at 2 ng/ml or supplement combination G [d3–5: TGFβ 1 10 ng/ml, BMP-2 10 ng/ml; d3–32: BMP-2 10 ng/ml, ascorbic acid 50 μg/ml and insulin 1 μg/ml]. Green fluorescing chondrocytes in both cultures appeared either organized in clusters or scattered as seen in figure 3A . The Kramer protocol gave a 7.26 percent yield of chondrocytes, which was increased to 57.03% using our modified protocol (fig. 3B ). Figure 3 Quantification of chondrogenic yield by flow cytometry. (A, B) Genetically modified ES cells expressing GFP from the chondrocyte-specific aggrecan promotor. Green fluorescing chondrocytes appear as clusters of cells within the remaining cell population (A), but are also scattered in the entire population starting at day 28 of differentiation (B). (C) FACS analysis of ES-derived chondrocytes sorted by their GFP expression. Differentiation with BMP-2 only [2 ng/ml, d3–5] produced 7.26% GFP expressing cells compared to spontaneously differentiated controls, which contained very low levels of fluorescing cells (1.54%). Prolonged BMP-2 administration together with TGFβ 1 , ascorbic acid and insulin (supplement combination G, see table 1) raised chondrocyte outcome to 57.03%. Kinetic analysis of cartilage-specific gene expression during EB differentiation The degree of chondrogenic differentiation using BMP-2, TGFβ 1 , insulin and ascorbic acid supplementation was then examined over the 35-day culture period using quantitative RT-PCR of various cartilage matrix genes (fig. 4 ). Since the extent of changes in cartilage-specific genes seemed to be dependent on acute (d 3–5) or chronic (d 3–32) application of BMP-2 (supplement combination F versus supplement combination G), the stronger induction by chronic supplementation of BMP-2 was monitored throughout the culture duration. During the first three weeks of the BMP-2 induced differentiation, only minor changes in aggrecan, link protein or collagen type II expression could be detected. Starting with day 26 however, aggrecan expression was up-regulated significantly to 14-fold over control niveaus (P < 0.1). Reaching day 32, aggrecan, link protein and collagen type II A and B approached peak levels. Consistent with the aggrecan expression profile, link protein expression was found to be significantly up-regulated on days 24–27 (factor 6.3; P = 0.001). Increased values were detectable as early as day 16. On day 26, the quantification of the collagen type IIA and B showed 11- and 24-fold increases respectively. On day 14, splice form B had already reached an 8-fold increase over control values. In contrast, during the early phases of differentiation, between day 5 and 20, collagen type IIA expression was continuously increased 2-fold. Transcripts for biglycan and decorin were detectable throughout all stages of chondrogenic differentiation. With the beginning of the maturation phase in the fourth week of culture (day 21–28), both genes were up-regulated 1.5–2 fold. Transcription factors scleraxis and Sox9 (fig. 4B ) showed a similar expression profile constantly over the entire culture period. This level did not change significantly during chondrogenic differentiation, but rather decreased compared to controls. Both were already transcribed in day 5 EBs, hallmarking their participation in early differentiation events, where lineage specificity is determined. Additionally, scleraxis transcripts were engaged in later phases of development between days 16–19. Sox9 expression was also increased between days 9–11 and between days 25–27. In conclusion, cultures supplemented with BMP-2, TGFβ 1 , insulin and ascorbic acid express mRNAs of a chondrogenic phenotype, whose expression was time-dependent. Figure 4 Expression of cartilage-specific genes in the course of EB differentiation induced by BMP-2, TGFβ 1 , insulin and ascorbic acid (supplement combination G, table 1). (A) Aggrecan, collagen type II A and B and link protein, biglycan and decorin. (B) Scleraxis and Sox9. Results show induction factors of expression obtained by quantitative RT-PCR and normalized to GAPDH expression in comparison to corresponding spontaneously differentiated controls, which were set as 1 (mean ± standard deviation, n = 3 independent experiments, 20 EBs each). ES-derived chondrocytes undergo hypertrophy and mineralize During embryo development endochondral ossification occurs in two steps: chondrocytes arise after mesenchymal condensation and become hypertrophic, characterized by expression of collagen type X, and calcification. To test whether this was true for the ES-derived chondrocytes generated with our protocol, we assayed for the Ca 2+ content of the cultures, a measure for mineralization (fig. 5A ). Calcium was increased 1.2-fold in cells that were supplemented with BMP-2, TGFβ 1 , insulin and ascorbic acid compared to controls (P < 0.01). Previously, we have described the induction of mineralization in osteoblasts derived from ES cells with vitamin D 3 , which reach maturity at day 32 of culture [ 18 , 27 ]. Here, we observed that the level of mineralization was considerably higher in direct osteoblast differentiation [11.57 mg/dl] compared to indirect differentiation using our improved chondrocyte protocol [3.22 mg/dl]. We observed, however, that the level of calcium found in VD 3 induced ES-derived osteoblasts could be rescued in the chondrocytes by adding VD 3 on day 20 of differentiation, a time when chondrocytes could be morphologically identified in the cultures (11.18 mg/dl, P < 0.001). As shown by quantitative RT-PCR, VD 3 treated ES cell derived chondrocytes could alter their expression profile to that of d32 ES cell derived osteoblasts (fig. 5B ). However, expression of osteocalcin and bone sialoprotein was less than in VD 3 rescued chondrocyte cultures than in VD 3 osteoblast cultures. Interestingly, chondrocyte-specific genes could not be detected in VD 3 osteoblasts and waned in VD 3 rescued chondrocytes. As we have shown previously, ES-derived mineralized osteocalcin expressing osteoblasts can be identified as black appearing cells in phase contrast microscopy [ 18 ]. Figure 5C shows these black cells in the VD 3 treated cultures. No such cells are visible in control cultures or in ES-derived chondrocyte differentiations. However, by adding VD 3 back in at day 20 to the chondrocytes, mineralization can be detected, although the localization pattern is slightly different. These observations support the hypothesis that VD 3 induces intramembraneous bone formation directly from mesenchymal progenitors while BMP-2 controls endochondral bone formation. Figure 5 Degree of osteogenesis in cultures treated with chondroinducing medium (BMP, supplement combination G), osteoinducing medium (VD 3 as described in ref. 18) and control ES medium (ctrl). Osteoblast phenotype and mineralization could be increased in chondrocyte cultures by adding VD 3 to chondroinducing supplements at day 20 (BMP+VD 3 ). (A) Extent of mineralization as measured by Ca 2+ content. P < 0.01; P = 0.001. (B) RT-PCR for chondrocyte-specific and osteoblast-specific marker genes as well as Collagen X as a marker for the hypertrophic state of chondrocytes. (C) Morphology as seen in phase contrast. Black appearing cells were identified as being mineralized osteocalcin expressing osteoblasts previously [18]. No such cells can be seen in control (ctrl) and chondrocyte (BMP) but in osteoblast (VD 3 ) and rescued osteoblast cultures (BMP+VD 3 ). It has been shown by others that the BMP-2 induced alteration in cell fate is both concentration- and time-dependent [ 28 ]. Lower concentrations of BMP-2 support chondrogenesis whereas higher concentrations promote osteogenesis. In ES cells, BMP-2 at concentrations of 2 ng/ml, 10 ng/ml and 100 ng/ml did not increase the expression of bone markers with the exception of osteocalcin and osteopontin, which were significantly increased as shown by quantitative RT-PCR (fig. 6 ). In combination with VD 3 (given on days 5–30) however, alkaline phosphatase and Cbfa1 were also significantly up-regulated above controls (5.4- and 4-2-fold, respectively, P < 0.001). As we noted earlier, BMP-2 induced osteogenesis with or without VD 3 supplementation did not meet the levels that were attained by VD 3 alone, but the late addition of VD 3 on day 20 rescued bone-specific gene expression arguing for an involvement for BMP-2 in endochondral bone formation. Figure 6 Dependence of genes expressed in cartilage tissue on BMP-2 with and without vitamin D 3 (VD 3 ). All cultures contained β-glycerophosphate [10 mM] and ascorbic acid [50 μg/ml]. Mean ± SD of independent triplicates was quantified by qPCR and is normalized to GAPDH expression. Controls were set as 1. P < 0,01; **P = 0,001. OCN = osteocalcin, BSP = bone sialoprotein, Cbfa1 = Core binding factor alpha, ALP = alkaline phosphatase, OPN = osteopontin, ONT = osteonectin, Col1 = Collagen type I. Concentrations of BMP-2 up to 100 ng/ml do not lead to osteoblast-specific expression levels that are reached with VD 3 only. 100 ng/ml BMP-2 plus VD 3 given together early during differentiation can significantly up-regulate osteoblast genes. However, when VD 3 is administered later (d20) in combination with BMP-2, osteoblast gene expression is almost restored. Induction of adipocyte differentiation During treatment of EBs with various chondrocyte differentiation inducing factors, we noticed accumulation of lipid droplets, which could not be found in untreated controls. Those droplets could indeed be characterized as lipid-containing by means of Oil-Red-O staining (fig. 7A ). Quantitative real-time PCR analysis revealed a slight up-regulation of adipocyte-specific genes (ADD1, aP2, C/EBPα, GLUT-4, LPL, PPARγ and SCD1) on day 30 in most of the supplement combinations used for induction of chondrocyte differentiation (fig. 7B,C ). Supplement combination G, which was the most successful for inducing chondrocyte differentiation suppressed adipocyte differentiation. GLUT-4 was most up-regulated in combinations B and C, whereas combinations E and B induced a increase in LPL expression compared to our chondrocyte differentiation protocol. Figure 7 Characterization of ES-derived adipocytes. (A) Oil-Red-O staining of lipid droplets in ES-derived adipocytes. Bar = 42 μm. (B, C) Influence of BMP-2 alone and in combination with TGFβ 1 , insulin and ascorbic acid (see table 1) on expression of adipocyte-specific genes. Quantitative RT-PCR was performed on 30 day old EBs. Expression of genes of interest was normalized to GAPDH and compared to untreated controls. Values show means ± standard deviations on n = 3 independent experiments. P < 0.1; P < 0.01; *P = 0.001. Discussion In this study, we demonstrate an improved method for driving ES cells to a cartilaginous fate when stimulated with BMP-2 and TGFβ 1 . In the early phase of differentiation, BMP-2 operates by directing differentiation towards the cartilage lineage and acting on chondroprogenitors. During later differentiation adding mineralizing agents can trigger hypertrophy and mineralization of the ES-derived chondrocytes. In the embryo, maturation of chondrocytes in the process of endochondral ossification follows a timely regulated developmental program, whereby cellular stages can be delimitated molecularly. During ES cell differentiation into chondrocytes, developmental processes follow the same pattern, as judged by the gene expression patterns observed. Chondrocytes, osteoblasts and adipocytes are thought to arise from the same mesenchymal progenitor. Based on our previous observations around VD 3 -induced osteogenesis, we have already hypothesized that during the first 5 days of differentiation mesodermal progenitors develop, which then are susceptible to the VD 3 treatment [ 18 ]. Treatment of the cultures with TGFβ 1 at days 3–5 may augment the number of mesenchymal progenitors, as TGFβ 1 is thought to inhibit the proliferation of most cells, but to stimulate some mesenchymal cells such as osteoblasts and chondrocytes [ 8 ]. It was not surprising to see adipocytes develop in many of our cultures, as they represent another member of the TGFβ 1 promoted mesodermal lineage. The gain of adipose characteristics in culture is hallmarked by: a) the appearance of cytoplasmic lipid droplets, b) the acquisition of insulin sensitivity with regard to glucose uptake (GLUT4) and c) the expression and secretion of numerous bioactive molecules [ 5 ]. All of these characteristics of in vivo adipogenesis were met in the EBs treated with BMP-2, TGFβ 1 , insulin and ascorbic acid. Indeed, a future challenge for improving adipogenic cultures will be the discovery of regulatory pathways in adipogenesis. Once identified, such pathways may be antagonized in order to enhance ES differentiation into chondrocytes. During the revision of this paper, a study was published describing that the overexpression of the sox triad, namely Sox9, Sox5 and Sox6, markedly increased chondrocyte marker gene expression (collagen type 2, aggrecan) in ES cells within 3 days [ 29 ]. Here, TGFβ 1 , BMP-2, IGF-1 and FGF-2 had no affect on the immediate regulation of these genes. In our differentiation system, Sox9 is elevated very early during differentiation at day 5 (data not shown) underlining its role as an early controller of chondrogenesis. We have shown here that BMP-2 regulates later processes in cartilage development. Marker gene expression levels reached upon overexpression of the Sox trio do not meet the levels we have observed in this study, suggesting that Sox9 may be necessary but not sufficient do direct all progenitors to the chondrocytic lineage. Earlier this year, another study portrayed chondrogenesis in ES cells using encapsulation in alginate [ 26 ]. The usage of a 3D culture system led to an increase in Col II and aggrecan expression of about 1.5- to 2-fold above the regular 2D system of plating EBs. Comparing those values to the 80- to 90-fold up-regulation described here, it is clear that three-dimensional signals also need to be incorporated into chondrocyte differentiations to increase differentiation efficiency. Articular cartilage has been refractory to repair following degeneration. Despite its limited capacity to self-repair, cartilage is replete with cells capable of undergoing mitotic division [ 30 ]. Current hypotheses suggest that cells may be constrained by their ECM, thus preventing expansion and differentiation or that there is a limit in the bioactive molecules, which support chondrogenesis [ 31 , 32 ]. Engineering bone or cartilage usually requires the handling of autologous cells. Cells are released from the ECM using collagenase and hyaluronidase. However, the small number of available progenitors, within the tissue can be problematic [ 33 ]. Moreover, the ability of these harvested cells to proliferate is limited in the elderly, where most degenerative joint disorders occur. The outcome of conventional surgical treatment including joint resurfacing or biological autografts has been unsatisfactory following long-term evaluation [ 34 ]. This failure is caused by insufficient repair resulting in the formation of mechanically inadequate resident fibrocartilage. These disappointing results and the limited therapeutic opportunities have led investigators to focus on more appropriate bioregenerative tissue engineering approaches, which could be specifically tailored for a patient's needs. Pluripotent ES cells are now being contemplated as a new cell source for tissue engineering since they are the most multifaceted cells amongst all stem cells. While their pluripotency offers a huge potential for cell therapies directed against a wide spectrum of degenerative diseases that are ineffectively treated by traditional approaches, it also poses challenge for controlling developmental fate in the recipient. Understanding the developmental pathways regulating ES cell differentiation, will enable the creation of a cell source that can be manipulated to correct for a particular defect. This study was designed to improve upon the number of differentiated cells needed for the in vitro development of functional cartilage as this represents a first critical step in applying ES cells clinically. Additionally, the presented in vitro model of chondrogenesis may be useful for in vitro embryotoxicity tests [ 27 ] as also serve as a new tool in identifying molecular pathways that underlie chondrogenesis. Compared to other in vitro models of chondrogenesis, this model incorporates all steps of development beginning with a pluripotent, uncommitted cell and thus might further our understanding of normally unamendable stages of development. Conclusions This study was particularly designed to improve chondrocyte yields from ES cells by investigating the effect of higher BMP-2 concentrations and the complementary effects of TGFβ 1 , insulin and ascorbic acid. By examining gene expression responses and cell sorting for GFP-expressing chondrocytes, we demonstrate that using a higher concentration of BMP-2 in combination with appropriate co-factors, we can significantly enhance ES cell derived chondrogenesis compared to known protocols. In addition, we document that ES-derived chondrocytes behave naturally as they undergo hypertrophy. We show that EBs supplemented with BMP-2 also result in a small amount of adipogenesis in vitro . This observation is consistent with knowledge that adipocytes, chondrocytes and osteoblasts arise from the same mesenchymal ancestor. Accordingly, in the future, it will be necessary to understand how to discriminate these populations for cytotherapeutic applications. At this time, the presented in vitro model allows the study of mechanisms involved in BMP-2 induced chondrogenesis, osteogenesis and adipogenesis. Methods Cell culture and differentiation of embryonic stem cells Cells of the mouse ES cell line D3 (American Type Culture Collection, Rockville, Maryland, USA) were kept in permanent culture as described [ 35 , 18 ] with the additive Leukemia Inhibitory Factor (1000 U/ml, Gibco Life Technologies, Karlsruhe, Germany). Differentiation was initiated in hanging drops. Cells condensed to form EBs, which were transferred on day 3 into suspension culture. At day 5, EBs were plated into 24-well tissue culture primaria plates (Falcon, Heidelberg, Germany). The effects of BMP-2 [2 or 10 ng/ml], TGFβ 1 [2 ng/ml], insulin [1 μg/ml] and ascorbic acid [50 μg/ml] in various combinations on the differentiation of chondrocytes were examined. Medium was changed every second day. Alcian blue staining Proteoglycans secreted by ES-cell derived chondrocytes were stained with Alcian blue. Cultures were fixed in 2.5% glutaraldehyde, 25 mM sodium acetate, 0.4 M MgCl 2 containing 0.05% Alcian blue for 48 h. Wash steps in 3% acetic acid, 3% acetic acid/25% ethanol and 3% acetic acid/50% ethanol reduced unspecific binding of the dye. Metachromatic test with 1.9-dimethylmethylenblue Proteoglycan content of differentiated cultures was determined with the DMMB-assay [ 36 ]. Proteoglycans were extracted in 4 M guanidin-HCl/0.05 M sodium acetate (pH 5.8) containing 100 mM 6-amino-caproic acid, 10 mM EDTA, 5 mM benzamidine/HCl, 10 mM N-ethylmaleimide, 0.4 mM pepstatin, 1 mM PMSF and 1 μg/ml soy bean trypsin inhibitor for 48 h at 4°C. Non-completely digested cells were separated from the lysate by centrifugation. Lysate was mixed with DMMB reagent (0.16% w/v DMMB in 0.2% formic acid containing 2 mg/ml sodium formate, pH 3.5) and changes in absorption were detected at 535 nm in a spectrophotometer. Concentration of proteoglycans in samples was read against a standard curve of chondroitin sulfate C. Immunofluorescence Embryoid bodies were differentiated to the chondrocyte lineage with BMP-2, TGFβ 1 , insulin and ascorbic acid and fixed on day 32 with ice-cold methanol:aceton (7:3) at -20°C for 10 min. For staining with anti-collagen type II (Chemicon, MAB 8887) cultures were digested with pepsin for 15 min at 37°C. Staining with anti-adult proteoglycan (Chemicon MAB 2015) was performed after a 1 h digest with chondroitinase ABC at room temperature. Cells were overlaid with the appropriate dilution of the first antibodies in PBS, 10% FCS at 4°C over night. The corresponding secondary antibody, an AlexaFluor 488 goat anti-mouse IgG (H+L), F(ab') 2 fragment (A-11006, Molecular Probes, Leiden, The Netherlands) was incubated with the cells for 2 h at room temperature. Cultures were observed in a Leica Fluovert FU fluorescent microscope (Leitz, Wetzlar) with an excitation wavelength of 495 and an emission wavelength of 519 nm. RNA isolation, RT-PCR and real-time quantitative RT-PCR Total RNA was isolated from 20 EBs per probe using the RNeasy Midi Kit (Qiagen, Hilden, Germany) according to the manufacturer's instructions with on-column DNase I digestion. The amount of RNA was determined using the RiboGreen™ RNA quantitation reagent and kit (Molecular Probes). 275 ng total EB RNA was used as a template for cDNA synthesis with Superscript II (Invitrogen, Paisley, Scotland) as described [ 27 , 35 ] in a total reaction volume of 25 μl. 5 μl aliquots of the first strand reaction were used for amplification performed with specific primers (see table 2 ). Primer sequences for osteoblast-specific genes and PCR conditions have been described previously [ 18 ]. PCR products were visualized on 3% agarose gels containing 0.1 μg/ml ethidium bromide. Quantitative real-time PCR analysis was performed in an ABI Prism ® 7700 Sequence Detector. The accumulation of reaction products during PCR was monitored by measuring the increase in fluorescence caused by the binding of SYBR ® Green (Applied Biosystems, Perkin Elmer, Weiterstadt, Germany) to double-stranded DNA. Expression analysis of collagen type II A and B was done in the same PCR run with two differently labelled probes specific for either the A isoform (5'-CGAGATCCCCTTCGGAGAGTGCTGT-3'/VIC) or the B isoform (5'-CCAGGATGCCCGAAAATTAGGGCCAA-3'/FAM) [ 27 ]. Reaction mixtures were set up as suggested by the manufacturer containing either SYBR Green or probes for collagen type II. Following a 10 min Taq Polymerase activating step at 95°C, the reactions were cycled by denaturing at 94°C for 30 s and annealing and elongation for 30 s at the corresponding temperatures (table 2 ). Target gene C T -values were standardized against GAPDH expression and induction of expression in treated EBs was normalized to control EBs. Primer sequences for murine GAPDH were 5'-GCACAGTCAAGGCCGAGAAT-3' and 5'-GCCTTCTCCATGGTGGTGAA-3' (T a = 60°C). Table 2 Sequences and annealing temperatures for the primer sets used in RT-PCR Gene Primer sequence T m in °C Chondrocyte-specific Aggrecan 5'-GATCTGGCATGAGAGAGGCG-3' 5'-GCCACGGTGCCCTTTTTAC-3' 61 Collagen type II 5'-GCTGCTGACGCTGCTCATC-3' 5'-GGTTCTCCTTTCTGCCCCTT-3' 60 Collagen type X 5'-CAAGCCAGGCTATGGAAGTC-3' 5'-AGCTGGGCCAATATCTCCTT-3' 60 Link protein 5'-TTCTGGGCTATGACCGCTG-3' 5'-AGCGCCTTCTTGGTCGAGA-3' 60 Biglycan 5'-CATGACAACCGTATCCGCAA-3' 5'-ATTCCCGCCCATCTCAATG-3' 60 Decorin 5'-ATGACCCTGACAATCCCCTG-3' 5'-CCCAGATCAGAACACTGCACC-3' 60 Scleraxis 5'-GGACCGCAAGCTCTCCAAG-3' 5'-ACCCACCAGCAGCACATTG-3' 62 Sox9 5'-GCAGACCAGTACCCGCATCT-3' 5'-CTCGCTCTCGTTCAGCAGC-3' 62 Adipocyte-specific ADD1 5'-CAGTGACTCTGAGCCCGACA-3' 5'-ATGCCTCGGCTATGTGAAGG-3' 61 PPARγ 5'-ATCATCTACACGATGCTGGCC-3' 5'-CTCCCTGGTCATGAATCCTTG-3' 59 SCD1 5'-ACACCATGGCGTTCCAAAAT-3' 5'-CGGCGTGTGTTTCTGAGAACT-3' 61 C/EBPα 5'-CGCAAGAGCCGAGATAAAGC-3' 5'-GCGGTCATTGTCACTGGTCA-3' 60 GLUT-4 5'-ATGGCTGTCGCTGGTTTCTC-3' 5'-ACCCATAGCATCCGCAACAT-3' 59 AP2 5'-TGATGCCTTTGTGGGAACCT-3' 5'-GCAAAGCCCACTCCCACTT-3' 58 acrp30 5'-AAGAAGGACAAGGCCGTTCTC-3' 5'-GAGGAGCACAGAGCCAGAGG-3' 60 LPL 5'-CCAATGGAGGCACTTTCCAG-3' 5'-CCACGTCTCCGAGTCCTCTC-3' 60 Forward primer sequence is depicted from the 5' to the 3' end followed by the reverse primer. The appropriate annealing temperature for each set is given as T m in °C. Quantification of chondrocyte yield by FACS The GFP expressing plasmid pEGFP-1 (Clontech) under the control of the Aggrecan promotor (kind gift of John R. Matyas) was stably transfected into D3 ES cells using Invitrogen's Effectene system followed by neomycin selection. Genomic integration of the reporter was analyzed by RT-PCR with specific GFP primers (data not shown). ES cells were differentiated along the chondrocyte lineage, trypsinized into a single cell suspension and subjected to fluorescence-activated cell sorting on day 32 using a FACS Calibur instrument and the CellQuest software from Becton Dickinson (Germany). Ten thousand events were registered per sample and analysis of whole cells was performed using appropriate scatter gates to avoid cellular debris and aggregates. Oil-Red-O staining Cells were washed with PBS and without any fixation directly overlaid with Oil-red-O working solution (0.18% Oil-Red-O dye/60% propanol) After a staining period of 15 minutes, ES cell cultures were rinsed in distilled water until desired colour was achieved. Statistical analysis Data analysis was performed with ONE-WAY ANOVA on n = 3 experiments using SigmaStat version 2.03 (SPSS Inc., San Rafael, CA, USA). List of abbreviations ADD1 Adipocyte determination- and differentiation-dependent factor 1 aP2 Fatty acid-binding protein BMP Bone Morphogenetic Protein Cbfa1 Core binding factor alpha 1 C/EBP CCAAT/enhancer-binding protein Col Collagen DMMB Dimethylmethylenblue ECM extracellular matrix ES embryonic stem EB embryoid body FCS Fetal calf serum FGF Fibroblast growth factor GAPDH Glyceraldehyde-3-phosphate dehydrogenase GLUT4 Glucose transporter 4 IGF Insulin-like growth factor PPAR Peroxisome proliferator-activated receptor SCD Steroyl CoA desaturase Sox Sry-related high mobility group box TGFβ Transforming Growth Factor beta VD 3 Vitamin D 3 Authors' contributions NZN carried out cell culture, biochemical and molecular studies and drafted the manuscript. GK, DER and HJA participated in its design and coordination. All authors read and approved the final manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC548146.xml
535807	Inconsistent self-reported mammography history: Findings from the National Population Health Survey longitudinal cohort	Background Self-reported information has commonly been used to monitor mammography utilization across populations and time periods. However, longitudinal investigations regarding the prevalence and determinants of inconsistent responses over time and the impact of such responses on population screening estimates are lacking. Methods Based on longitudinal panel data for a representative cohort of Canadian women aged 40+ years (n = 3,537) assessed in the 1994–95 (baseline) and 1996–97 (follow-up) National Population Health Survey (NPHS), we examined the prevalence of inconsistent self-reports of mammography utilization. Logistic regression models were used to estimate the associations between women's baseline sociodemographic and health characteristics and 2 types of inconsistent responses: (i) baseline reports of ever use which were subsequently contradicted by follow-up reports of never use; and (ii) baseline reports of never use which were contradicted by follow-up reports of use prior to 1994–95. Results Among women who reported having a mammogram at baseline, 5.9% (95% confidence interval (CI): 4.6–7.3%) reported at follow-up that they had never had one. Multivariate logistic regression analyses showed that women with such inconsistent responses were more often outside target age groups, from low income households and less likely to report hormone replacement therapy and Pap smear use. Among women reporting never use at baseline and ever use at follow-up, 17.4% (95%CI: 11.7–23.1%) reported their most recent mammogram as occurring prior to 1994–95 (baseline) and such responses were more common among women aged 70+ years and those in poorer health. Conclusions Women with inconsistent responses of type (i), i.e., ever users at baseline but never users at follow-up, appeared to exhibit characteristics typical of never users of mammography screening. Although limited by sample size, our preliminary analyses suggest that type (ii) responses are more likely to be the result of recall bias due to competing morbidity and age. Inconsistent responses, if removed from the analyses, may be a greater source of loss to follow-up than deaths/institutionalization or item non-response.	Background In the absence of organized screening, self-report is often the only means to monitor mammography utilization and to investigate trends in uptake at the population level [ 1 ]. In Canada, mammographic screening occurs both within organized programs and opportunistically through routine medical practice [ 2 ]. The validity of mammography self-report has previously been studied primarily using convenience samples. Despite differences in methodology, design and population characteristics, studies from a variety of settings have found that self-reports of mammography use are valid provided women are not required to precisely recall the timing of a previous mammogram [ 3 - 14 ]. Women generally tend to underestimate the time elapsed since their most recent mammogram by an average of three months or more, though overestimation can also occur [ 4 - 6 , 8 - 13 , 15 - 17 ]. Greater discrepancies in recall may occur when the mammogram took place longer ago [ 15 , 16 ], though contrary evidence also exists [ 11 ]. With some exceptions [ 4 , 6 , 8 , 9 , 13 , 16 , 17 ], studies have not been designed to assess false negative reporting due to the challenge of verifying historical data from multiple service providers. Those that have attempted to validate non-use have indicated that women are unlikely to deny having had a mammogram when indeed they have had one [ 4 , 6 , 8 , 9 , 13 , 17 ], though false-negative self-reports are not always negligible [ 16 ]. However, the opposite is often true; women tend to report having had mammograms which are not verified against medical records [ 3 , 4 , 6 , 7 , 9 - 11 , 13 - 16 , 18 - 21 ], a particular problem in groups with low screening prevalence [ 20 ]. Valid reporting of mammography use has been found to be unrelated to various health behaviors and perceptions, socioeconomic, demographic, and questionnaire administration factors in some studies [ 4 , 10 , 12 , 13 ]. However, others provide some evidence that age, ethnicity, education, employment status, family history of breast cancer, recency of the mammogram, and the regularity with which women receive mammograms do affect self-report accuracy [ 7 , 10 - 13 , 17 , 22 ]. Few evaluations of the reliability of mammography self-report are reported in the literature. Excellent test-retest reliability for having ever had a mammogram was reported in interviews conducted 1 week, 6–30 days, or 6–8 months after an initial interview, while reliability within the past year varied from excellent to good [ 14 , 23 , 24 ]. In a socioeconomically advantaged group of women aged 50–75 followed annually for 3 years, 98 percent provided logically consistent responses to a question on ever/never use of mammography [ 25 ]. Self-reports of ever use have been shown to be more reliable among Caucasian women and those with higher income and education [ 24 ]. However, in this study, date of last mammogram was not as reliably reported 6–8 months after initial testing [ 24 ]. Based on data from the longitudinal panel of the National Population Health Survey (NPHS), the present study examines the prevalence and determinants of inconsistent self-reports of mammography utilization among Canadian women aged 40 years and older and quantifies the extent that inconsistent self-reports of mammography use contribute to biased estimates of mammography utilization and uptake. To our knowledge, this is one of the first studies of mammography utilization to provide specific longitudinal data on the determinants of inconsistent responses over time and the impact of such responses on population screening estimates. Methods The National Population Health Survey (NPHS) is a survey of the Canadian household population. Initiated in 1994–95 and repeated biennially, it is a split panel survey, combining repeated cross-sectional components with the longitudinal follow-up of a panel of respondents. A representative sample of Canadian household residents aged 12 and older from all ten provinces was sampled using a multistage probability design with stratification and clustering at various stages. The overall response rate for the baseline 1994–95 survey was 89 percent with a 96 percent response rate for the selected panel respondent [ 26 ]. On follow-up to the baseline survey two years later, 94 percent of the panel members responded [ 26 ]. Further details of the sampling procedures, design, data collection and response rates are published elsewhere [ 26 , 27 ]. This study evaluated data from longitudinal panel respondents of the 1994–95 (baseline) and 1996–97 (follow-up) waves of the NPHS to examine inconsistencies in reported mammography utilization among women aged 40+ years at first contact. Questions about mammography use were administered to female respondents through a personal interview conducted in 1994–95 and repeated by telephone approximately two years later. In both survey years, women were asked the identical question: "Have you ever had a mammogram, that is, a breast x-ray?". Those with positive responses were further probed for the time and reason of their most recent mammogram. All women provided their own health-related information; no proxy responses were allowed. Analyses were restricted to women aged 40 and older (at baseline) who participated in the first two waves of the NPHS and consented to share their information with federal and provincial governments. Two types of inconsistent responses were assessed: (i) baseline reports of ever use which were contradicted by follow-up reports of never use; and (ii) baseline reports of never use which were contradicted by follow-up reports of use prior to 1994–95. Multivariate logistic regression techniques were used to evaluate the associations between women's baseline sociodemographic and health characteristics and type (i) inconsistent responses. Variables significant at p ≤ .05 in age-adjusted analyses were eligible for entry in the multivariate logistic models. Sample size constraints permitted only simple bivariate, rather than multivariate exploration of factors associated with reports reflecting inconsistent timing of most recent use at follow-up (type (ii) response). Estimates were weighted to reflect baseline population characteristics. To account for stratification and clustering in the NPHS sampling design, 95% confidence intervals for parameter estimates were calculated using exact standard errors generated through bootstrap re-sampling methods [ 28 ]. All statistical analyses were conducted using SAS. Results (i) Inconsistent ever/never utilization Of the 3,535 women aged 40+ years who responded to the ever/never mammography question in both survey waves (Figure 1 : 2 women with missing data regarding timing of mammogram were excluded), four percent (95% CI: 3.1–4.9) reported having had a mammogram at baseline and subsequently, on follow-up reported never having had a mammogram (Table 1 ). Among women who reported having had a mammogram at baseline, 5.9% (95%CI: 4.6–7.3) reported never use at follow-up (estimate not shown). The majority of women with inconsistent responses (64.4%, 95% CI: 54.4–74.4) reported receiving a recent (i.e., <2 years ago) mammogram at baseline and most (85.6%, 95% CI: 78.2–93.1) reported that the mammogram was done as part of a regular check up (Table 1 ). It should be noted that the percentage estimates in Table 1 have been weighted according to 1994/95 population characteristics whereas the frequency data represent actual numbers of women surveyed. Table 2 presents the estimated adjusted odds ratios (95% CIs) of inconsistent ever/never responses associated with women's baseline sociodemographic and health characteristics. Among women reporting ever use at baseline (1994–95), those reporting never use in 1996–97 were significantly more likely to be outside the target age group for screening (50–69), to have lower income, to have not used hormone replacement therapy in the past month and to have never had a Pap test, after adjusting for relevant covariates. Women with lower education levels were also more likely to report such inconsistent responses between baseline and follow-up although education failed to remain a significant predictor in the multivariate model. Other variables considered but not found to be significantly associated with this outcome were rural/urban residence, place of birth, languages spoken, marital status and other social support indices and having a regular physician. (ii) Inconsistent timing Follow-up interviews were completed, on average, 1.98 years from the baseline survey (range 1.19–3.01 years). Of the 293 women who reported never use at baseline and ever use at follow-up, 17.4 percent (95%CI: 11.7–23.1) reported a time for their most recent mammogram at follow-up that was inconsistent with never use at baseline. Despite baseline reports that they had never had a mammogram, approximately half of these women reported having had a mammogram at least 5 years ago. Although limited by small numbers, determinants of such inconsistent responses were assessed with simple bivariate analyses. Inconsistencies in timing occurred more often in older women. Compared to women aged 50–69, those 70 and older were more likely to report (at follow-up) that their most recent mammogram had occurred prior to 1994–95, despite a report of never use at baseline (OR = 6.96, 95%CI: 2.42–20.0). Women reporting fair or poor self-rated health were also more likely to report a time for their most recent mammogram at follow-up that was inconsistent with never use at baseline (OR = 2.44, 95% CI: 0.99–6.05). Impact of inconsistent reporting on uptake estimates Depending on how inconsistent responses are handled, different measures of use and uptake of mammography may be obtained. The lack of a gold standard such as a medical chart for validation makes the choice of a corrective measure unclear. If inconsistent ever/never responses are included in the analysis unchanged, 67.3 percent (95% CI: 65.1–69.5%) of women would be classified as ever having had a mammogram in 1994–95 while 71.7 percent (95% CI: 69.6–73.7%) would be classified as ever users in 1996–97. Conversely, if it is assumed that inconsistent ever/never responses represent false-positive responses at baseline (an assumption supported by our study findings), the 1994–95 prevalence estimate becomes 63.3 percent (95% CI: 61.0–65.6%), demonstrating an absolute increase in mammography use of 8.4 percent (95% CI: 7.1–9.6%) by this cohort of women by 1996–97. Discussion Although a limited number of studies have assessed the reliability of mammography self-report [ 23 - 25 ], detailed evaluations have not been conducted for population-based longitudinal surveys. In this study, reliability could not be assessed, per se, as women's status of never having had a mammogram could normally be expected to change over a two year span. However, by examining inconsistencies in responses expected to remain constant and in responses regarding logical timing of mammography use, it is possible to examine potential concerns regarding response reliability and recall bias, respectively. In longitudinal studies, inconsistent data removed during data cleaning can yield significant losses, and may lead to bias, depending on the amount of attrition at each time point and the magnitude of the differences between those retained in the panel and those lost by such attrition [ 29 ]. Longitudinal studies of health must be acutely aware of causes of attrition because losses accumulate over survey waves [ 30 ]. Although direct comparison with our sample was not possible due to the expectation that behavior might have changed in a 2-year time span, earlier findings from longitudinal studies of fairly affluent [ 25 ] and low-income [ 24 ] populations and a population-based study [ 23 ], indicated that women reliably report having ever had a mammogram with estimated reliability measured by Cohen's kappa ranging from 0.82–0.87 [ 23 , 24 ]. Our finding that 4 percent (95% CI: 3.1–4.9 percent) of the women participating in the second wave of this longitudinal study inconsistently reported ever having had a previous mammogram was surprisingly high. Previous studies have found that initial use refuted on subsequent interviews occurred in 2–2.9 percent of respondents [ 24 , 25 ]. Our analyses of factors associated with inconsistent ever/never responses indicate that women reporting ever having had a mammogram at baseline but never use at follow-up exhibited many of the sociodemographic and health behaviour characteristics (e.g., lower income, outside age groups targeted for screening, non-users of Pap screening and hormone replacement therapy) commonly observed among non-participants in mammography screening in previous studies [ 31 - 38 ]. Such findings provide support for the assumption that the 1994–95 response of ever use is more likely erroneous. Additional factors (e.g. being born in an Asian country) previously associated with non-use of mammography [ 31 , 33 ] also showed a positive association with providing an inconsistent response; however, small numbers resulted in high variability once clustering and stratification were taken into account and precluded further analysis of this variable. Validation studies also provide support for our assumption that inconsistent ever/never responses (as assessed in the present longitudinal panel) are most likely to be explained by false-positive responses at baseline. Women are more likely to falsely claim having had a mammogram, than not having one [ 4 , 6 , 8 , 9 , 13 , 17 , 39 , 40 ]. The majority of women in our study with inconsistent ever/never responses also indicated (at baseline) that their most recent mammogram had occurred within the last two years, a finding consistent with past research [ 10 ]. Imputation, suggested as a remedy for item non-response [ 30 ], may equally be used to deal with inconsistencies with evidence, in this situation, favoring treatment of women's earlier responses as false-positive. Among women who reported never use at baseline and ever use at follow-up, approximately 17.4 percent reported a time for their most recent mammogram (at follow-up) that was inconsistent with never use at baseline. If respondents truly initiated mammography use subsequent to the baseline interview, they overestimated the time elapsed since their mammogram. Such a finding is relatively inconsistent with previous studies that have generally found that women tend to underestimate the time since their last mammogram [ 4 - 6 , 8 - 13 , 15 , 17 ]. Although McGovern et al. and Caplan et al. found reverse-telescoping in approximately 9 percent of their samples [ 13 , 16 ], only 8 percent of women in this group miscalculated by more than 1 year [ 16 ]. In the present study, women reporting inconsistent timing were older and in poorer health, suggesting that competing health events may have interfered with accurate recall. Unfortunately, no gold standard was available to assess the validity of the responses women generated. Thus, the proportion of consistent responses that may actually represent invalid responses is unclear. Nor was it possible to distinguish errors in recall of timing from false reports of mammography utilization. Further, the reasons for providing inconsistent responses can only be inferred. The possibility of data entry errors is remote. Although data entry checks for consistency between survey cycles were not included among the comprehensive quality control strategies implemented by Statistics Canada, computer assisted interviewing was used by highly trained interviewers. Also, data entry errors would be expected to occur randomly and not disproportionately among women outside of the target age range or with relatively lower socioeconomic status, as observed in the present study. However, several plausible explanations for the inconsistencies exist, including survey methodology changes and deliberate or inadvertent provision of inaccurate responses by the respondent. We cannot exclude the possibility that interview changes (from a baseline personal interview to a telephone follow-up) prompted women who reported ever use in 1994/95 to alter their response in 1996–97. The NPHS used an initial personal interview to foster a good long-term relationship with the panel representative, but the cost and logistics of traveling to different regions was prohibitive. Therefore, unless the respondent objected or had no phone, future interviews (including the 1996–97 cycle) were conducted by telephone [ 26 ]. The need to maintain study procedures over time in longitudinal studies has been stressed [ 41 ], but the impact of altering the interview method on the NPHS results has not been investigated [ 26 ]. Sensitive questions may be answered more truthfully by phone. Editing of survey responses by the respondent may occur. Social desirability and a tendency to give positive responses are possible sources of over-reporting [ 3 , 11 ]. Such biases remain largely unexplored with respect to mammography use. Cognitive research has implicated comprehension as a barrier to providing valid, reliable survey responses. Several researchers employ the lead-in question 'Have you ever heard of a mammogram, that is a breast x-ray?' to identify comprehension difficulties [ 3 , 16 , 21 , 42 ] but this was not done in the NPHS where women were directly asked if they had ever had a mammogram. One study using focus group testing and in-depth interviews showed that despite some confusion between mammography and breast exams, women generally understood what mammography was [ 11 ], a finding further supported by one population-based survey [ 21 ]. However other investigations suggest this is not uniformly so [ 16 , 42 , 43 ]. One possibility cited is that confusion with other tests such as chest x-rays may lead to over-reporting of mammography [ 3 ]. There are several possible explanations for the higher rate of inconsistent responses observed in the present study. The focus of the NPHS questionnaire was not limited to preventive practices nor was it designed for reliability testing. The length and comprehensiveness of the NPHS may have contributed to greater respondent fatigue. Also, the longer time interval between surveys, relative to that observed in other studies, may have contributed to instability in responses. A more favourable level of concordance among responses may be obtained by studies that apply eligibility criteria that ensure more accurate responses (e.g. by having the respondent recall where she had her mammogram to validate her ever/never use) [ 6 , 10 , 15 ]. Finally, the overall response rate of the NPHS was higher than in comparable studies, so it potentially included a more difficult-to-reach population, less able to provide accurate responses. The inconsistent data reported here, if removed from longitudinal analyses, could yield losses to follow-up equivalent to or greater than other sources of attrition (e.g., deaths, institutionalization, non-response) over the planned 20 year course of the NPHS. The 1998–99 NPHS included probing questions to reduce inconsistencies and it should alleviate many of the problems evident here [ 44 ]. However, probes were not designed to address the larger problem of reverse-telescoping observed among some respondents in the NPHS longitudinal cohort. Incorporation of women's previous responses in subsequent interviews to avoid telescoping and stimulate recall can be used to minimize such inconsistencies [ 30 ]. A recent critical review of the accuracy of self-reported health behaviors, including mammography, provides further suggestions for enhancing the accuracy of such data [ 45 ]. Conclusions In summary, inconsistent responses represent a challenge to longitudinal, population-based evaluations of breast screening practices. Losses from inconsistent data regarding mammography participation are not negligible and may contribute to inaccurate estimates of mammography uptake. Women reporting inconsistent ever/never use in the present study displayed characteristics typical of never users, favoring treatment of women's baseline responses as false-positive. Inconsistent responses regarding the timing of recent mammography practices, however, may be primarily related to the impact of age and competing morbidity on recall. Competing interests The authors declare that they have no competing interests. Authors' contributions CB and CM contributed to the initial and revised analyses of the NPHS panel data, the interpretation of the results and the initial drafting/editing of the manuscript. CB and JS were responsible for the data linkage with Statistics Canada (for the NPHS Share File) and for the revised bootstrap analyses. JS also contributed to the interpretation of the results and editing of the manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC535807.xml
549516	Prognostic indicators in peritoneal carcinomatosis from gastrointestinal cancer	Peritoneal carcinomatosis from gastrointestinal cancer has new treatment options for surgical management. The approach uses cytoreductive surgery which combines peritonectomy and visceral resection in an effort to remove all visible cancer within the abdomen and pelvis. Then the peritoneal cavity is flooded with chemotherapy solution in an attempt to eradicate residual disease. In order to select patients for this approach the quantitative prognostic indicators for carcinomatosis were reviewed, compared and contrasted. Prognostic indicators to be used to select patients for this aggressive approach at the initiation of surgery and after completion of cytoreduction were studied. Four quantitative assessments to be used at the time of abdominal exploration were the Gilly staging, Japanese gastric cancer P score, peritoneal cancer index (PCI), and the simplified peritoneal cancer index (SPCI). All have value with the PCI being the most validated and most precise. Preoperative assessments include the tumor histopathology and the prior surgical score. The completeness of cytoreduction score is an assessment of residual disease after a maximal surgical effort. An opportunity for long-term survival following treatment for carcinomatosis requires a complete cytoreduction in all reports for gastrointestinal cancer. Quantitative prognostic indicators need to be knowledgeably employed when patients with carcinomatosis are being treated. Improved patient selection with greater benefit and reduced morbidity and mortality should result.	I. Introduction Peritoneal carcinomatosis has always been regarded as a terminal condition. It is present in 10 to 30% of patients with gastrointestinal cancer at the time of their initial surgery and is a frequent finding in patients who develop recurrent cancer. Important natural history studies establish a 6-month median survival in this group of patients [ 1 - 3 ]. Recent multicenter phase II and a single phase III study evaluating the usefulness of cytoreductive surgery and perioperative intraperitoneal chemotherapy are promising [ 4 , 5 ]. Patient selection is of utmost importance in optimizing the results of treatment and excluding patients who will not benefit from a high morbidity and potentially life threatening therapy. Quantitative prognostic indicators are to serve as guidelines in the selection of treatments to maximize benefits of therapy and to exclude patients who have little or no chance to improve. They are of greatest utility in high risk and costly management protocols. Requirements of a useful quantitative prognostic indicator include reproducibility, prediction of survivorship, and assessment of morbidity and mortality. The goal is to establish management protocols that standardize the decision making process for multiple caregivers. General surgery has used quantitative prognostic indicators in the past with established benefit to patient care. Examples of quantitative prognostic indicators currently in use include Ranson's criteria, which estimates the risk of life threatening complication or death in patients with acute pancreatitis; and, the Child-Pugh score for liver cirrhosis, which evaluates the severity of liver disease correlating grades with one- and two-year survival. Currently, there are several clinical assessments at many different institutions in use for the evaluation of carcinomatosis (see Table 1 ). Our goal in this manuscript is to critically discuss these quantitative prognostic indicators. Collaborative studies between institutions would be greatly facilitated with standardized clinical tools for management of carcinomatosis from gastrointestinal cancer. Table 1 Quantitative prognostic indicators currently in use in patients with carcinomatosis. Tumor histopathology Intraoperative assessment of the extent of carcinomatosis at time of surgical exploration • Gilly peritoneal carcinomatosis staging • Carcinomatosis staging by the Japanese Research Society for Gastric Cancer • Peritoneal Cancer Index (PCI) • Dutch Simplified Peritoneal Carcinomatosis Index (SPCI) CT PCI Prior Surgical Score Completeness of Cytoreduction Score II. Histopathology In patients with carcinomatosis from gastrointestinal cancer, invasive implants are disseminated within the peritoneal cavity. However, in two conditions the biological aggressiveness of the disease will have a broad spectrum. These two diseases are mucinous appendiceal malignancies (oftentimes clinically designated pseudomyxoma peritonei syndrome) and peritoneal mesothelioma. In these diseases a non-invasive process may be widely disseminated on the peritoneal surfaces. The biological aggressiveness of the malignancy can be estimated by the pathologist in a knowledgeable histologic review of multiple specimens. For pseudomyxoma peritonei syndrome, the histologic classification described by Ronnett and colleagues has been most widely utilized [ 6 ]. Histopathologic examination categorizes the disease process into disseminated peritoneal adenomucinosis (DPAM), peritoneal mucinous carcinoma (PMCA) or a hybrid type. Disseminated Peritoneal Adenomucinosis This is a minimally invasive disease, and therefore more likely to be completely removed by cytoreduction using peritonectomy. The histology of DPAM shows a bland single layer of epithelium that surrounds lobules of mucin. There are no signet rings and there is minimal atypia. Invasion of the structures upon which tumor accumulates does not occur. The primary site for DPAM is an appendiceal adenoma which has minimally invaded the wall of the appendix. Usually, the widespread intraperitoneal dissemination of mucinous tumor is caused by a rupture of the lumen of the appendix from pressure built up by the malignant mucocele. Figure 1 shows the typical disruption of the wall of the appendix by tumor. Figure 2 presents the histologic character of DPAM. Figure 1 Right colon, terminal ileum and mucocele of the appendix. This appendix is greatly dilated; the end has ruptured releasing mucus and adenomatous epithelial cells into the free peritoneal cavity. Figure 2 Histopathology of disseminated peritoneal adenomucinosis (DPAM). (H+E × 200) Peritoneal Mucinous Adenocarcinoma This is an invasive disease in which the mucinous cancer cells show invasion into surrounding tissues. Sometimes, the signet ring morphology or lymph node metastases are present. The cancer cells will be found in multiple layers surrounding the mucinous tumor globules. There is loss of nuclear polarity and atypia is common. The quantity of mucus may be variable from one patient to the other. However, the PMCA histology may be associated with very large amounts of mucoid ascites fluid. Therefore, it is categorized as pseudomyxoma peritonei syndrome but with an aggressive tumor histology. Figure 3 provides an example of PMCA. Figure 3 Histopathology of peritoneal mucinous adenomucinosis (PMCA). (H+E × 700) Hybrid Type Disease In the hybrid type of mucinous carcinomatosis the field of view for the pathologist shows 95% or more DPAM. PMCA is present but in 5% or less of the total field of view (figure 4 ). If there is more than 5% PMCA the histology is no longer hybrid type but designated as PMCA. Figure 4 Histopathology of hybrid type mucinous appendiceal malignancy. (H+E × 100) Not surprisingly, the observation has been made by numerous groups that the non-invasive mucinous tumors (DPAM and hybrid type) are amenable to complete cytoreduction. Therefore more definitive treatment and improved survival using the combined approach is expected with DPAM and hybrid type [ 7 ]. The histologic and clinical differences between the different types of mucinous appendiceal and other gastrointestinal mucinous tumors are shown in Table 2 . Table 2 Histopathologic features of epithelial mucinous tumors of appendiceal, colonic, and small bowel origin are designated as disseminated peritoneal adenomucinosis (DPAM) and peritoneal mucinous carcinomatosis (PMCA). Features DPAM PMCA Primary site Appendix Appendix, colon, small intestine Primary diagnosis Mucinous adenoma usually in a mucocoele Mucinous adenocarcinoma Surgical appearance Mucinous tumors and mucinous ascites with redistribution Carcinomatosis with variable amounts of mucinous ascites, redistribution is prominent with large volume of ascites Peritoneal tumor • Cellularity Scant Moderate to abundant • Morphology Abundant extracellular mucin containing simple to focally proliferative mucinous epithelium. There is a single layer of cells Moderate to abundant extracellular mucin containing extensively proliferative mucinous epithelium or mucinous glands, clusters of cells, or individual cells consistent with carcinoma • Cytologic atypia Minimal Moderate to marked • Mitotic activity Rare Infrequent to frequent Lymph node involvement Almost never Moderate Liver metastases Almost never Very infrequent Parenchymal organ invasion Rare (except ovary) Frequent Hybrid type tumors show less than 5% of PMCA within DPAM. Mucinous carcinomas are divided into three grades by maintenance or loss of glandular architecture. III. Intraperitoneal Assessment of the extent of Carcinomatosis The quantitation of tumor found at the time of surgical exploration of the abdomen has proven to be of value in assessment of prognosis and treatment planning. Four different assessments have been published. They are listed in Table 1 . Gilly Peritoneal Carcinomatosis Staging The Gilly peritoneal carcinomatosis staging format was first described in Lyon in 1994 [ 8 ]. This prognostic tool takes into account the size of lesions found at operation (table 3 ). Two advantages of this system are simplicity and reproducibility. The utility of the Gilly staging device in survivorship prediction has been demonstrated in the multicentric prospective EVOCAPE study which gathered data from 370 patients with peritoneal carcinomatosis from non-gynecologic malignancies [ 2 ]. A significant difference was observed between stages 1 and 2 with a median survival of 6 months and stages 3 and 4 whose median survival was 3 months. The Gilly carcinomatosis staging has also been validated in patients having combined treatment for carcinomatosis [ 9 ]. Table 3 Gilly peritoneal carcinomatosis staging. Stage Peritoneal carcinomatosis description Stage 0 No macroscopic disease Stage 1 Malignant implants less than 5 mm in diameter Localized in one part of the abdomen Stage 2 Diffuse to the whole abdomen Stage 3 Malignant implants 5 mm to 2 cm Stage 4 Large malignant nodules (more than 2 cm) Although the Gilly system has been used for almost a decade with acceptable prognostic value, there are some criticisms regarding this system. First, it should not be designated a "staging system" because patients can only be staged once in the course of their disease at the time of diagnosis of the primary malignancy. Usually, a TNM staging system is appropriate. The system might better be called the Gilly prognostic index for carcinomatosis. A second weakness of the Gilly prognostic index concerns a failure to quantitate distribution of peritoneal surface implants in the stage 3 and 4 categories. Carcinomatosis confined to one portion of the abdomen may carry an excellent prognosis even if the localized tumor implants are of large size. If group III and group IV nodules by size are diffuse throughout the whole abdomen, certainly a much different prognosis would occur. A definitive assessment of not only the size of the nodules but also the distribution of carcinomatosis is necessary for the most accurate assessment of prognosis. The Japanese have proposed a quantitation of carcinomatosis that is very simple, has been frequently applied, and has been validated for gastric malignancy. For the original staging a "P factor" is indicated for gastric cancer patients. P-0 means that no carcinomatosis was seen by the surgeon or could be established at the time of surgery. It would currently include patients who are cytology positive for gastric cancer cells. P-1 indicates implants immediately adjacent to the stomach and above the transverse colon. P-2 indicates scattered implants within the abdomen but not of great number. P-3 indicates numerous implants throughout the abdomen and pelvis. This staging system can also be applied to patients who have carcinomatosis with recurrent gastric cancer. A major deficit of this staging system is its inability to accurately locate the carcinomatosis. Also, it has no size assessment of the cancerous implants. Although the P factor has been of great value historically in the management of primary gastric cancer as peritonectomy and intraperitoneal chemotherapy are used for treatment of carcinomatosis, a more precise prognostic assessment is needed to manage gastric cancer peritoneal seeding. Peritoneal Cancer Index The Peritoneal Cancer Index (PCI), like the other carcinomatosis assessments, is determined at the time of surgical exploration of the abdomen and pelvis. With invasive cancer it serves as an estimate of probability of complete cytoreduction and has been found to be an accurate assessment of survival when cytoreductive surgery and perioperative intraperitoneal chemotherapy are used as treatment [ 10 ]. The PCI quantitatively combines the distribution of tumor throughout 13 abdominopelvic regions with a lesion size score. Two transverse and two sagittal planes divide the abdomen into 9 regions. The upper transverse plane is located at the lowest aspect of the costal margin, and the lower transverse plane is placed at the anterior superior iliac spine. The sagittal planes divide the abdomen into three equal sectors. The lines define 9 regions, which are numbered in a clockwise direction with 0 at the umbilicus and 1 defining the space beneath the right hemidiaphragm. Regions 9 through 12 divide the small bowel into upper and lower jejunum and upper and lower ileum (Figure 5 ). To make the PCI tool more quantitative and reproducible, each region is not only defined by the surface landmarks as previously described, but can also be defined by the anatomic structures found in each region (Table 4 ). Figure 5 Peritoneal cancer index (PCI). Two transverse planes and two sagittal planes divide the abdomen into 9 regions. The upper transverse plane is located at the lowest aspect of the costal margin and the lower transverse plane is placed at the anterior superior iliac spine. The sagittal planes divide the abdomen into three equal sectors. The lines define the nine regions which are numbered in a clockwise direction with 0 at the umbilicus and 1 defining the space beneath the right hemidiaphragm. Regions 9–12 divide the small bowel. Lesion size score is determined after complete lysis of all adhesions and the complete inspection of all parietal and visceral peritoneal surfaces. It refers to the greatest diameter of tumor implants that are distributed on the peritoneal surfaces. Primary tumors or localized recurrences at the primary site that can be removed definitively are excluded from the lesion size assessment. If there is confluence of disease matting abdominal or pelvic structures together, this is automatically scored as L-3 even if it is a thin confluence of cancerous implants. Table 4 Anatomic structures involved in the 13 abdominopelvic regions of the peritoneal cancer index (PCI). Regions Anatomic structures 0 Central Midline abdominal incision – entire greater omentum – transverse colon 1 Right upper Superior surface of the right lobe of the liver – undersurface of the right hemidiaphragm – right retro hepatic space 2 Epigastrium Epigastric fat pad – left lobe of the liver – lesser omentum – falciform ligament 3 Left upper Undersurface of the left hemidiaphragm – spleen – tail of pancreas – anterior and posterior surfaces of the stomach 4 Left flank Descending colon – left abdominal gutter 5 Left lower Pelvic sidewall lateral to the sigmoid colon – sigmoid colon 6 Pelvis Female internal genitalia with ovaries, tubes and uterus – bladder, Douglas pouch – rectosigmoid colon 7 Right lower Right pelvic sidewall – cecum – appendix 8 Right flank Right abdominal gutter – ascending colon 9 Upper jejunum 10 Lower jejunum 11 Upper ileum 12 Lower ileum The lesion size (LS) score is determined after complete lysis of all adhesions and complete inspection of all parietal and visceral peritonea surfaces within the abdominopelvic regions. LS-0 indicates no implants seen. LS-1 indicates implants less than 0.25 cm. LS-2 indicates implants between 0.25 and 2.5 cm. LS-3 indicates implants greater than 2.5 cm. It refers to the greatest diameter of tumor implants that are distributed on the peritoneal surfaces. Primary tumors or localized recurrences at the primary site that can be removed definitively are excluded from the assessment. If there is a confluence of disease matting abdominal or pelvic structures together, this is automatically scored as LS-3 even if it is a thin layer of cancerous implants. The lesion sizes are then summated for all abdominopelvic regions. The extent of the disease within all regions of the abdomen and pelvis is indicated by a numerical score from 0 to 39. In 1995, Sugarbaker and Jablonski published that the PCI was a meaningful assessment for colon cancer but not for mucinous appendiceal tumors [ 11 ]. Elias et al., found survival to be more favorable in those patients with carcinomatosis from colon cancer with a PCI score of less than 16 [ 12 ]. In a larger number of patients Sugarbaker and Chang established survivorship using the PCI [ 13 ]. Five-year survival was 50% in colon cancer patients with carcinomatosis with a PCI less than 10, 20% for 11–20 and 0% in those with a PCI score greater than 20 (Figure 6 ). Tentes and colleagues validated the PCI for ovarian cancer [ 14 ]. The PCI is not only useful as a prognostic indicator but also as a guide for sequential determinations of volume of carcinomatosis over time estimating the likelihood of a complete cytoreduction at re-operative surgery [ 15 ]. Figure 6 Peritoneal carcinomatosis from colon malignancy survival by peritoneal cancer index. (Modified from Reference 13) This quantitative prognostic indicator for colon carcinomatosis established that for patients scoring greater than 20, palliation is the goal of treatment. Currently, a PCI of greater than 20 is regarded as a relative contraindication to an elective intervention for carcinomatosis from colon cancer. It is associated with a low median survival, approximately the same as median survival without surgical intervention. In patients who have a PCI greater than 20, palliative surgery is indicated in order to alleviate symptoms or to prevent symptoms that may occur in the near future. In an asymptomatic patient with colon carcinomatosis cytoreductive surgery with intraperitoneal chemotherapy with cure as a goal of treatment is probably not indicated. An exception to the utility of the PCI is found in treating patients with pseudomyxoma peritonei and minimally aggressive mesothelioma. Because the disease is non-invasive, a PCI of 39 can be converted to 0 by cytoreductive surgery. There is a low probability of recurrence after complete cytoreduction with perioperative intraperitoneal chemotherapy and therefore the PCI has no prognostic implication [ 7 ]. Another caveat that must be observed when using the PCI occurs in cases in which a low PCI score is recorded in the presence of invasive cancer at a crucial anatomic site. For example, at exploration one may find invasive tumor in and around the common bile duct with little disease elsewhere. Even thought the PCI is low, a complete cytoreduction may not be possible. In these cases, invasive cancer at a crucial anatomic site places the patient into the same category as would systemic metastasis in the lungs or bone. Only palliative surgery is indicated if residual disease post-cytoreduction will be present. Simplified Peritoneal Cancer Index The Simplified Peritoneal Cancer Index (SPCI) was established at the Netherlands Cancer Institute and has been used for colorectal and appendieal cancer staging (Table 5 ). This tool has prognostic implication for survival following cytoreductive surgery and hyperthermic intraperitoneal chemotherapy [ 16 ]. Table 5 Simplified Peritoneal Cancer Index ◆ Tumor is recorded as: indent="1" • Large (> 5 cm) indent="1" • Moderate (1–5 cm) indent="1" • Small (< 1 cm) indent="1" • None ◆ Seven abdominal regions: indent="1" • I: pelvis indent="1" • II: right lower abdomen indent="1" • III: greater omentum, transverse colon and spleen indent="1" • IV: right subdiaphragmatic area indent="1" • V: left subdiaphragmatic area indent="1" • VI: subhepatic and lesser omental area indent="1" • VII: small bowel and small bowel mesentery Verwaal and colleagues have provided important information regarding the relationship of the Simplified Peritoneal Cancer Index and the incidence of complications in patients who receive combined treatment [ 17 ]. In their review of the toxicity of combined treatment, complications increased when the cancer index recorded involvement of more than five regions (p = 0.044). Also, if the patient had recurrent colon cancer (as opposed to carcinomatosis with primary cancer) or if there was an incomplete cytoreduction, the incidence of complications was significantly higher. Verwaal et al., established that the peritoneal cancer index quantitated not only the survival outcome of these patients but also the expected morbidity and mortality of the combined treatment [ 16 ]. There are marked similarities between the SPCI and the PCI. Both the anatomic distribution of the tumor masses and the size of the tumor masses within each abdominal region are indicated. In the PCI, there are 13 anatomic sites designated by a diagram; in the Dutch SPCI, there are 7 anatomic regions designated by anatomic site. In both systems the volume of tumor in each region is to be scored quantitatively. Some shortcomings of the SPCI could be formulated. First, the epigastric region, very important in determining the completeness of cytoreduction in some diseases is not designated separately. Disease above the stomach in the lesser omental region may cause the cytoreduction to be incomplete [ 15 ]. A second major criticism of the Dutch SPCI concerns their misuse of their own tool. In their recent publications they perform a survival analysis by SPCI and a toxicity assessment by the SPCI. However, only the involvement of regions 0–7 was indicated. No tumor size in the regions was indicated [ 16 , 17 ]. Prior Surgical Score An accepted fact regarding cancer treatment is that the optimal treatment with the highest cure rate, the greatest preservation of function, and the lowest morbidity and mortality is the initial treatment. In the management of carcinomatosis the extent of prior resection before definitive cytoreduction with intraperitoneal chemotherapy has a negative impact on the survival. This occurs because of the cancer cell entrapment phenomenon. Surgery opens tissue planes whose raw surface is a favored site for cancer cell adherence, vascularization and progression. In the use of combined treatment for carcinomatosis, the non-traumatized peritoneal surface is the body's first line of defense against carcinomatosis. Cancer progression deep to peritoneal surfaces, especially disease imbedded in scar, is difficult or impossible to remove by peritonectomy or to eradicate by intraperitoneal chemotherapy. The prior surgical score (PSS) quantitates the extent of surgery prior to definitive combined treatment. It shows that the greater the surgery the poorer the results of carcinomatosis treatment. The assessment uses a diagram similar to that for PCI but excludes abdominopelvic regions 9–12. For a PSS of 0 no prior surgery or only a biopsy was performed; PSS of 1 indicates one region with prior surgery; PSS-2 indicates 2 to 5 regions previously dissected; PSS-3 indicates more than 5 regions previously dissected. This is equivalent to a prior attempt at complete cytoreduction but in the absence of perioperative intraperitoneal chemotherapy. In appendiceal cancer patients with a prior surgical score of 0–2, the survival using combined treatment was 70% at 5 years; with a prior surgical score of 3, the 5-year survival was 51% (p = 0.001) [ 18 ]. Completeness of Cytoreduction Score The Completeness of Cytoreduction Score functions as a major prognostic indicator for the survival in peritoneal mesothelima, colon cancer with carcinomatosis, gastric cancer with carcinomatosis and sarcomatosis [ 7 ]. It is to be assessed after cytoreductive surgery is completed. Complete cytoreduction (CC-0 or CC-1) or incomplete (CC-2 or CC-3) are determined. A CC-0 is apparent when there is no peritoneal seeding visualized within the operative field. CC-1 indicates nodules persisting after cytoreduction less than 2.5 cm. CC-2 has nodules between 2.5 and 5 cm, whereas a CC-3 indicates nodules greater than 5 cm or a confluence of unresectable tumor nodule at any site within the abdomen or pelvis. The CC-1 tumor nodule size is thought to be penetrable by intracavitary chemotherapy and is, therefore, designated as complete cytoreduction if perioperative intraperitoneal chemotherapy is used. Sugarbaker and colleagues found that prognosis can be estimated by completeness of cytoreduction. For colon cancer as shown in Figure 7 , there is a 40% chance of survival at 5 years in those who undergo complete cytoreduction versus 0% survival in the incomplete category [ 7 , 13 ]. Numerous other groups have confirmed the complete cytoreduction as a requirement for survival after treatment of carcinomatosis from appendiceal, colorectal and gastric cancer [ 4 , 9 , 11 , 12 , 16 - 19 ]. Figure 7 Peritoneal carcinomatosis from colon malignancy survival by cytoreduction. (Modified from Reference 13) Although no formal statement in the literature is available, it is thought that the definition of complete vs. incomplete cytoreduction varies with the histologic type of the malignancy. For example, mucinous tumors by diffusion are well penetrated with intraperitoneal chemotherapy solutions. With minimally invasive mucinous tumors such as pseudomyxoma peritonei, complete cytoreduction may occur in the combined treatment plan with tumor nodules up to a full centimeter in size. In contrast, hard fibrotic non-mucinous colon cancer is poorly penetrated by chemotherapy solution. Only cytoreduction down to no visible evidence of disease would be expected to result in long-term survival with a sclerotic malignant process. Also, some cancers may be remarkably more responsive to chemotherapy than others. This is likely the case with a majority of ovarian cancers. Their complete response to systemic chemotherapy is also frequently seen with intraperitoneal chemotherapy solutions or a bidirectional (intraperitoneal combined with intravenous chemotherapy) approach. In both these situations the definition of a complete cytoreduction scored by a CC-1 designation would vary with the clinical situation. Computerized Tomographic PCI The preoperative CT is an excellent tool in locating and quantifying mucinous adenocarcinoma within the peritoneal cavity [ 20 ]. Unfortunately, with intestinal histologic type of colon cancer the accuracy of the CT is considerably reduced [ 21 ]. However, for mucinous carcinomatosis CT scanning is an accurate prognostic indicator of the possibility of resectability. It may show segmental obstruction of the small bowel or tumor nodules greater than 5 cm on small bowel. Patients who have both of these findings have a likelihood of less than 5% of complete cytoreduction. Obstructed segments of bowel signal an invasive character of malignancy on small bowl surfaces that would be unlikely to be completely cytoreduced. Large tumor nodules on small bowel or its mesentery are unlikely to be adequately cytoreduced without visceral resection. There are some special demands on CT scanning if the radiologic examination is to be optimized. Bowel loops cut in cross section are often indistinguishable from cancer nodules. Only if maximal oral contrast using a barium sulfate compound is utilized to prepare the patient for this examination can the greatest accuracy and the greatest prognostic implications of the examination be realized. Another technical requirement is the imaging of solid tumor layered out on the peritoneal surfaces. Unless there is maximal intravenous contrast with a 60 to 120 second delay after contrast infusion will the confluence of malignancy as a thin layer on the peritoneum be imaged. In some patients, the solid tumor, or semisolid tumor may be distributed to appear as ascites on abdominal and pelvic CT. Much to the surgeon's dismay, upon opening the abdomen, a solid tumor mass filling the abdomen and pelvis and causing adherence of small bowel and small bowel mesentery will be revealed. In this situation, not even palliative surgery can be safely performed. In patients who clinically have a firm abdomen and in whom the surgeon suspects large volume of solid tumor, an ultrasound examination may be required in order to confirm an ascitic versus a solid component of the abdominal and pelvic malignancy. If ultrasound shows that there is only minimal or no ascites and that the large volume of tumor is solid or semisolid, surgical interventions are not beneficial. It is better to determine the nature of the carcinomatosis radiologically than at the time of a major surgical exploration. Conclusion Quantitative prognostic indicators are of value in management of peritoneal surface malignancy from gastrointestinal cancer. Preoperative CT PCI, intraoperative PCI and postresection CC score have all been reported valuable. As one knowledgeable applies these tests, proper selection of patients for combined treatment may increase benefit and decrease morbidity and mortality.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC549516.xml
514611	Interval exercise versus continuous exercise in patients with moderate to severe chronic obstructive pulmonary disease – study protocol for a randomised controlled trial [ISRCTN11611768]	Background Physical exercise has become a cornerstone of management of chronic obstructive pulmonary disease (COPD) because it leads to clinically relevant improvements of exercise capacity and health-related quality of life (HRQL). Despite the scarcity of randomised trials directly comparing exercise protocols, current guidelines recommend high intensity continuous exercise for lower extremities as the probably most effective exercise modality. However, for patients admitted to inpatient respiratory rehabilitation programmes, it is often difficult to initiate such an exercise programme because they are severely limited by dyspnoea and leg fatigue and therefore unable to perform continuous exercise at higher intensities and for periods longer than 30 minutes. Interval exercise may be an attractive alternative for these COPD patients because it allows high intensity exercise with recovery periods. The aim of this study is to assess if interval exercise compared to high intensity continuous exercise is not of inferior effectiveness in terms of HRQL and exercise capacity improvements but associated with better exercise tolerance in patients with moderate to severe COPD at the beginning of a respiratory rehabilitation. Methods/Design We will assign patients with moderately severe to severe COPD to either continuous exercise or interval exercise using a stratified randomisation. Patients will follow 12–15 exercise sessions during a comprehensive inpatient respiratory rehabilitation. Primary end point for effectiveness is HRQL as measured by the Chronic Respiratory Questionnaire (CRQ) two weeks after the end of rehabilitation and secondary endpoints include additional clinical outcomes such as functional exercise capacity, other HRQL measures, patients' experience of physical exercise as well as physiological measures of the effects of physical exercise such as cardiopulmonary exercise testing. Including expected drop-outs, we will need 52 patients per group to show differences corresponding to the minimal clinically important difference of the CRQ. Outcome assessors and investigators involved in data analysis will be blinded to group assignment until analyses have been carried out. Discussion Clinicians and the scientific community need evidence on the benefits and tolerance of exercise protocols available in clinical practice. The proposed trial will provide important and needed data on interval and continuous exercise for decision making in clinical practice.	Background Impaired exercise capacity, dyspnoea and reduced health-related quality of life (HRQL) are common complaints of patients with chronic obstructive pulmonary disease (COPD). A major exercise-limiting factor in COPD is peripheral muscle dysfunction characterised by atrophic muscles and reduced fatigue resistance due morphological and metabolic alterations of peripheral muscles[ 1 ] As much as 70% of COPD patients may be affected by peripheral muscle dysfunction.[ 2 ] Respiratory rehabilitation with physical exercise improves exercise capacity and HRQL.[ 3 ] Although physical exercise is a mandatory component of respiratory rehabilitation programmes[ 4 , 5 ], there is an ongoing debate about what type of exercise at which exercise intensity patients should perform.[ 1 , 6 ] There is substantial variation in exercise protocols used in practice[ 7 ] as well as in clinical trials[ 3 ]. Current guidelines recommend continuous exercise at high intensity for lower extremities[ 4 , 5 ] because a study indicated that high intensity may be more effective than low or moderate intensity.[ 8 ] However, data on high intensity continuous exercise come from a trial that included 19 patients with mild COPD who were able to exercise for 45 minutes five times per week during an outpatient programme.[ 8 ] For patients who need to be admitted to inpatient programmes because of more severe COPD and/or unstable health state, it is difficult to perform high intensity exercise and exercise sessions longer than 30 minutes because they are limited by dyspnoea and leg fatigue. Less than 20% may be able to sustain high intensity continuous exercise throughout the whole rehabilitation programme[ 9 ] To find a realistically tolerable exercise programme for these patients, who often initiate exercise programmes for the first time, is challenging. A solution to this dilemma may represent interval exercise[ 6 ] where patients exercise alternatively at high intensity and at low intensity, which allows short periods of recovery. Consequently, interval training may be better tolerated than high intensity continuous training. In addition, patients may be able to achieve a greater training load during the relatively short exercise sessions they can sustain. Patients and clinicians will accept interval exercise to treat peripheral muscular dysfunction in COPD only if it is not of inferior effectiveness compared with continuous exercise and if it is indeed associated with better compliance resulting from less dyspnoea and leg fatigue during exercise. There is limited evidence from three randomised controlled trials comparing interval exercise and continuous exercise.[ 10 - 12 ] A summary of these trials can be found in table 1 . Table 1 Trials on interval exercise in patients with COPD Population Exercise protocols and rehabilitation program Main results Coppoolse 1999 [10] 21 stable male COPD patients (mean age 65 years, FEV1 36.8% predicted) Group 1 : CT ergometer cycling at 60% of Wmax Group 2 : IT ergometer cycling at 90% of Wmax (1 min) and 45% of Wmax (2 min) 3 days/week plus CT ergometer cycling at 60% of Wmax 2 days/week 8 weeks inpatient rehabilitation with 5 exercise sessions per week of 30 min. No additional physical exercise. Significant increase of V O2 and decrease of minute ventilation with CT but no changes with IT. Significant increase of Wmax and decrease of leg pain during exercise with IT but not with CT. Only significant differences between CT and IT for V O2 /Wmax favouring CT. 91% of patients with CT and 90% of patients with IT completed the exercise program. Vogiatzis 2002 [11] 45 stable COPD patients (62% males, (mean age 65 years, FEV1 34.1% predicted) Group 1 : CT ergometer cycling at 50% of Wmax weeks 1–4, at 60% weeks 5–8 and at 70% weeks 9–12 Group 2 : IT ergometer cycling at 100% of Wmax (30 sec) and 45% of Wmax (30 sec) weeks 1–4, at 120% weeks 5–8 and at 140% weeks 9–12 12 weeks outpatient rehabilitation with 2 exercise sessions per week of 40 min. No additional physical exercise. Significant improvements of CRQ scores and Wmax and reductions of minute ventilation during CWRT in both groups. No significant differences between groups. Attendance rate for exercise sessions 88% for CT and 90% for IT. Kaelin 2001 [12] 19 stable COPD patients (89% males) (mean age 67 years, FEV1 26.9% predicted) Group 1 : CT walking on stepper (70 steps/minute) or treadmill (1.5 miles/hour). Increase of 1 MET every 2 weeks Group 2 : IT walking on stepper (70 steps/minute) or treadmill (1.5 miles/hour) with active rest to ratio of 2:1. Increase of 1 MET every 2 weeks 6 weeks outpatient rehabilitation with 3 exercise sessions per week of 10–30 min. Additional resistance training and flexibility training. Larger improvements of 6-minute walking distance with IT (80 meters) compared with CT (39 meters). No data on compliance. CT = Continuous training; IT = Interval training; CRQ = Chronic Respiratory Questionnaire Wmax = Maximum exercise capacity, measured by usual incremental exercise test; CWRT = Constant work rate test; V O2 = Maximum oxygen consumption MET = Metabolic equivalent The studies indicated that both interval and continuous training improved exercise capacity, dyspnoea and HRQL and showed insignificant differences between interval and continuous exercise. However, insignificant differences do not allow concluding that interval or continuous are of clinically equivalent effectiveness[ 13 ] These trials were too small to show clinical equivalence or non-inferiority and they did not provide evidence on the tolerance of these two exercise modalities. From a methodological point of view, the trial had several shortcomings because, for example, they did not provide details on concealment of random allocation or blinding of outcome assessors. In addition, in the trial with an inpatient rehabilitation.[ 10 ], patients of the interval exercise group had a mixed intervention (3 days of interval and 2 days of continuous exercise per week) so that differences can hardly be attributed to different interventions if they are detected at all. The investigators did not use steep ramp tests to determine exercise loads but normal incremental exercise tests. For interval exercise, muscle strength and anaerobic capacity is relevant because of the short high intensity intervals, but this is not measured by normal incremental exercise tests. In addition, training load tolerated during interval exercise may be underestimated when normal incremental exercise tests are used.[ 14 ] Exercise tests to establish training intensity should consider the exercise mode. Meyer et al. studied interval exercise in several studies [ 14 - 16 ] in patients with chronic heart failure who show similar patterns of physical deconditioning in terms of clinical manifestation as well as morphological and metabolic abnormalities [ 17 - 19 ] Meyer et al. used a steep ramp test to determine short time muscular maximum exercise capacity[ 14 ], which reflects muscle strength and anaerobic capacity, both relevant for interval exercise. Meyer et al. also assessed different ratios of work/recovery phases (1:2; 1:4 and 1:6) and found that with these ratios and relatively short phases of high intensity exercise (10–30 seconds), lactate did not accumulate presumably because of lactate elimination during the recovery phases.[ 14 ] Interval exercise with these ratios was therefore recommended as high intensity aerobic exercise modes for patients who do not sustain continuous exercise. Because of the scarcity of evidence on the comparative effectiveness of interval exercise for COPD patients, additional trials are needed.[ 1 , 6 ] Our primary objective is to assess if interval exercise is not of inferior effectiveness compared to continuous exercise of high intensity to improve HRQL and exercise capacity in patients with moderately severe to severe COPD and the secondary objective is to evaluate if interval exercise is better tolerated by COPD patients. Methods Study design (see figure 1 ) Figure 1 Flow of the study from screening for eligible patients to the final outcome assessment. All consecutive patients admitted to a teaching rehabilitation clinic for an inpatient respiratory rehabilitation (Klinik Barmelweid, Barmelweid, Switzerland) will be assessed for study eligibility by senior staff physicians. If patients are deemed eligible after exercise testing, which is part of the usual rehabilitation program, senior physicians will inform patients about the study orally and in writing. If patients are willing to participate and provide written informed consent they will be randomly assigned to respiratory rehabilitation with either interval or high intensity continuous exercise. Both groups will perform 12–15 exercise sessions and follow the rest of the rehabilitation programme. Follow-up assessments will be done during, at the end of the rehabilitation programme as well as two and twelve weeks afterwards when patients are back in their home environment. The Ethikkommission of the Kantonsspital Aarau, Aargau, Switzerland, has approved the study protocol. Patients We defined the following in- and exclusion criteria: Patients with COPD as defined by FEV1/FVC < 70% predicted, FEV1 < 50 % predicted after bronchodilation, with or without chronic symptoms (cough, sputum production) corresponding to a GOLD (Global Initiative for Chronic Obstructive Lung Disease) stage III-IV[ 20 ] and German as first or daily language. Exclusion criteria are arrhythmia (atrial flutter and fibrillation, ventricular tachycardia, premature beats > 8 per minute), ischemia during exercise testing, clinically decompensated Cor pulmonale or heart failure, untreated neoplasia or neoplasia that needed treatment within the previous two years, lung surgery within the previous three months, orthopedic, rheumatologic, vascular or neurological disorders that inhibit ergometer training, gymnastic or guided walking tours, and patients unable to perform or complete the six-minute walk test or the incremental cycle test Randomisation A third party not involved in the conduction of the trial will provide online central randomisation (DatInf GmbH, Tuebingen, Germany). A computerised 'minimisation' procedure will be used to avoid chance baseline imbalances in prognostically important variables[ 21 ] Minimisation variables will be exercise capacity (< 300 or ≥ 300 meters in six-minute walk test), the presence of affective disorders (Hospital Anxiety Depression Scale scores < or ≥ 8), status of COPD (unstable COPD = In- or outpatient medical care in the last eight weeks due to exacerbation of COPD versus stable COPD = no in- or outpatient medical care in the last eight weeks due to exacerbation of COPD) and the need for oxygen at rest (yes = long term home oxygen therapy or paO 2 < 55 mmHg or no = paO 2 ≥ 55 mmHg). Every time the study coordinator has been informed about an enrolled patients, he will enter the randomisation web site[ 22 ] enter the patient data required for randomisation (patient identification and stratification variables) and obtain group allocation. The study coordinator will then inform responsible physical therapists about group allocation. No other medical staff will have knowledge about group allocation. The randomisation provider will also send an e-mail to the study coordinator for each randomised patient with details on randomisation. This will ensure correct verification of group allocation after data analysis. Separating patient enrolment and baseline assessments (physicians and physical therapists) from the randomisation procedure (study coordinator not involved in patient enrolment or rehabilitation programme) will ensure concealment of random allocation. Interventions The rehabilitation programme will start one day after baseline assessments and study enrolment. Exercise sessions and group lessons will take place five days a week and will consist of daily cycle ergometer training, breathing therapies (30 minutes per day), and guided walking of 15 to 30 minutes. Relaxation therapies (technique according to Jacobson) take place twice a week, patient education (information about COPD, coping strategies, inhalation techniques) three times a week, smoking cessation advice once a week or more if needed. Apart from physical exercise, the rehabilitation programme of approximately three weeks will be identical for both groups. Group performing continuous exercise The target workload for this group will be ≥ 70% of the maximum exercise capacity expressed in Watts and heart rate achieved during the incremental cycle ergometer test. Patients are usually not able to perform high intensity continuous exercise from the beginning and have to adapt to physical exercise. Physical therapists will increase training load as soon as possible to ≥ 70% of the maximum exercise capacity or as high as each individual patient tolerates. In each session (see figure 2 ), patients will have a warm-up period of two minutes at 20% of maximum exercise capacity, increase the exercise intensity within two minutes to the target intensity, exercise for 20 minutes at high intensity and then have a decreasing period of two minutes (gradual decrease from 70% to 0%). Pulse oxymetry will be used to supervise patients during exercise. If oxygen saturation falls below 90%, oxygen supplementation will be provided to maintain ≥ 90%. Figure 2 The upper graph shows the continuous exercise protocol for a patient who achieved a maximum exercise capacity of 100 Watts during a usual incremental exercise test. The lower graph shows the interval exercise protocol for a patient who achieved a short time muscular exercise capacity of 200 Watts during a steep ramp test. If patients cannot sustain the workload because of perceived dyspnoea or leg fatigue or because the heart rate exceeds the limits determined during exercise testing, physical therapists will let patients rest for one minute and then resume exercise. If patients have to rest more than twice per session, physical therapists will lower the workload by steps of 10% of baseline maximum exercise capacity. In turn, if patients or physical therapists consider the workload to be too low or if patients do not reach their target heart rate at 70% of the maximum exercise capacity, physical therapists will increase workload by steps of 10% of maximum exercise capacity until patients and physical therapists consider the workload to be appropriate or until the target heart rate is reached. Group performing interval exercise Patients assigned to this group will perform a steep ramp test to determine the short time muscular maximum exercise capacity.[ 14 ] The steep ramp test is an incremental cycle ergometer test where patients pedal unloaded for 2 minutes and then pedal with increments of 25 Watts every 10 seconds until they cannot maintain a pedaling frequency above 50 per minute or above the heart rate limit set by the normal incremental exercise test (which all patients perform irrespective of group assignment). With the steep ramp test measurement of muscular maximum exercise capacity is possible because the tests lasts only for 30 to 120 seconds and patients are not limited by symptoms. Measuring muscular maximum exercise capacity is important to set the exercise load for interval exercise because the high intensity interval requires also muscle strength beside overall exercise endurance.[ 14 ] Patients should improve both endurance and muscle strength during interval exercise because both are required in daily activities. We therefore chose a work/recovery ratio of 1:2 that prevents from high lactate accumulation.[ 14 ] From short time maximum exercise capacity, we will derive the initial work rate for interval exercise (in Watts), which is set at 50% of the short time muscular maximum exercise capacity as measured by the steep ramp test. This workload corresponds in Watts approximately to 90–100% of the workload as measured by the normal incremental exercise test.[ 16 ]. Patients will start with exercise the day after the steep ramp test. Patients will perform interval exercise for twelve to fifteen sessions with a cycle ergometer. In each session, they will have a warm up period of two minutes at 20% of the short time maximum exercise capacity (figure 2 ). Then they exercise for 20 minutes at high intensity intervals of 20 seconds at 50% and at low intensity intervals of 40 seconds at 20% of the short time maximum exercise capacity, i.e. with a work/recovery ratio of 1:2. Then patients have a slow down period of two minutes before completion of the training session. Pulse oxymetry will be used and oxygen supplementation will be provided as described above. If patients cannot sustain exercise intensity because the heart rate exceeds the limits determined after exercise testing or because of perceived dyspnoea or leg fatigue, physical therapists will let patients rest for one minute and then resume exercise. If patients have to rest more than twice per session, physical therapists will lower the workload from 50% of the short time maximum exercise capacity by steps of 10% while the length of intervals remains constant. They will increase the training load again as possible for the patient. In turn, if patients or physical therapists consider the workload to be too low, physical therapists will increase workload of the high intensity interval by steps of 10% until patients and physical therapists consider the workload to be appropriate while the length of intervals remains constant. Clinical outcome measures Chronic Respiratory Questionnaire (CRQ) We will use the CRQ[ 23 ] to measure HRQL changes during rehabilitation. The CRQ is a widely used disease-specific instrument to assess symptoms of COPD patients [ 24 - 26 ] We will use the self-administered German CRQ.[ 27 , 28 ] with standardised dyspnoea questions.[ 29 ] that we have developed and validated in earlier studies. Patients will complete the CRQ in the Klinik Barmelweid at baseline, at the end of the rehabilitation, two and 12 weeks thereafter when they have returned to their home environment. HADS (Hospital Anxiety Depression Score) Affective disorders are common in patients with COPD and contribute to reduced HRQL[ 30 ] The HADS has been developed to assess symptoms of anxiety and depression in patients with physical impairment.[ 31 ] There are seven items for each domain (anxiety and depression) with statements on emotions and emotional situations. Patients express their agreement with the statements on a scale from 0 to 3. Domain scores are calculated by summing up the scores for the seven domains resulting in scores from 0 (no depression or anxiety) to 21 (depression or anxiety very likely to be present). Scores ≥ 8 indicate that there is an increased probability for the presence of an affective disorder. We will use the validated self-administered German version of the HADS.[ 32 ] The HADS will be completed in the Klinik Barmelweid at baseline, at the end of the rehabilitation, two and 12 weeks thereafter when they have returned to their home environment. Feeling Thermometer (FT) We will use a validated visual analogue scale, the FT[ 33 ], an increasingly used instrument for a global estimate of the effect of interventions, including respiratory rehabilitation[ 29 , 34 ]. The FT is a visual analogue scale presented as a thermometer with 100 marked intervals. The worst (dead = 0) and best (perfect health = 100) health states are defined anchors and facilitate comparisons between individuals and groups.[ 35 ] We will ask patients to reflect in their score how they felt in the last 7 days. The FT will be completed in the Klinik Barmelweid at baseline, at the end of the rehabilitation, two and 12 weeks thereafter when they have returned to their home environment. Six-minute walk test We will use the six-minute walk test to assess functional exercise capacity according to established criteria.[ 36 ] At baseline, patients will perform the six-minute walk test twice one day apart. We will use the results of the second six-minute walk test because the first test tends to underestimate exercise capacity due to unfamiliarity with the test.[ 36 ]. At the end of the rehabilitation and two weeks after completion of rehabilitation programme patients will perform additional six-minute walk tests under supervision of physical therapists blinded to group assignment. In addition, we will use a paperboard with a modified Borg scale on from 0 to 10 with verbal labels for 0 (no dyspnoea at all), 1–5, 7 and 10 (maximal dyspnoea) to assess the intensity of perceived dyspnoea at the end of the six-minute walk test. Monitoring of exercise sessions Dyspnoea and leg pain during exercise In each session, we will use modified Borg scale as described above to assess the intensity of perceived dyspnoea and leg pain after five minutes of exercise and at the end of the exercise sessions. Subjective experience of exercise There is no instrument available to assess the subjective experience of COPD patients with exercise, which is likely to influence compliance. We have developed a questionnaire using established methodology[ 37 ] to assess how patients experienced the sessions. The questionnaire (Exercise Tolerance Questionnaire) consists of five questions addressing the exercise limitation by shortness of breath and difficulties with breathing, leg fatigue, fatigue in general and too high exercise load. In addition, one question asks patients how they experienced the exercise session in general (from very enjoyable to very unpleasant). The report on questionnaire development will be published elsewhere. Adherence to and tolerance of the prescribed cycle ergometer exercise Physical therapists will record for every cycle ergometer exercise session if patients exercised at all (yes or no), the performed workload (in Watts), if patients reached the target workload (in Watts), adjustments of workload and the requirement for oxygen. We define adherence to exercise as "full adherence" if patients follow at least 12 exercise sessions. We consider the training to be fully tolerated if patients are able to follow the exercise protocol for at least 9 exercise sessions for continuous exercise (taking into consideration the first three exercise session when patients increase training load up to 70% of maximum exercise capacity) and for at least 12 exercise sessions for interval exercise. Adverse events Previous trials of respiratory rehabilitation or physical exercise did not report any adverse events. Nevertheless, we will record any adverse events such as injuries, cardiac events or increase of respiratory symptoms as they happen during the inpatient respiratory rehabilitation. Cardiopulmonary exercise testing At baseline and end of the rehabilitation, all patients will perform an incremental cycle ergometer test to the limit of tolerance (symptom based) under the supervision of a senior physician blinded to group assignment. Patients pedal unloaded for three minutes to warm up at 20 Watts. Then exercise load is increased by 7.5 Watts per minute until patients have to stop because of dyspnoea, leg pain, or criteria for stop testing (atrial or ventricular tachycardia, ischemia, hypoxemia). At the limit of tolerance, we will draw capillary blood samples to measures lactate concentrations and we will set the maximum exercise capacity expressed by Watts. The upper limit for the heart rate during exercise is set as the heart rate measured by the electrocardiogram at the maximum exercise capacity. During testing, we will record gas exchange and ventilatory variables form calibrated signals derived from rapidly responding gas analyzers and a mass flow sensor. We will record breath by breath the following variables: Pulmonary oxygen uptake, pulmonary CO 2 output, minute ventilation, tidal volume and respiratory frequency. All patients will perform a steep ramp test at the beginning and end of the rehabilitation programme as described above. Additional data to be collected In order to characterize the patient included in the study, we will record their age, gender, lung function (FEV1, FEV1/FVC, diffusion capacity DLCO/VA, weight, height, smoking status at the beginning and end of rehabilitation as well as two and 12 weeks afterwards, duration of disease (time since diagnosis), co-morbidities such as hypertension, heart diseases, endocrine disorders, chronic rheumatological disorders and psychiatric disorders. Data analysis The randomisation code will not be broken (investigators remain blinded to group assignment and will receive only codes, such as group 1 and 2) until a draft of the manuscript has been written. The authors will write two versions with alternative possible allocation patterns to avoid bias in the interpretation of the results. This approach is methodologically rigorous and limits introduction of bias during the interpretation of the data that some of the authors might have.[ 38 ] We will submit the appropriate manuscript regardless of the results of unmasking. After agreement on the final versions of the two articles, we will break the randomisation code and we will submit the correct version of the manuscript. Effectiveness Our null hypothesis is that high intensity continuous exercise is of clinically superior effectiveness compared with interval exercise to improve HRQL (δ ≥ 0.5 in CRQ domains scores). The alternative hypothesis is that interval exercise is not of clinically inferior effectiveness compared with high intensity continuous exercise. In the primary analysis, we will calculate the raw difference and 95% confidence intervals between groups in the mean follow-up score for the CRQ domains 2 weeks after completion of the respiratory rehabilitation programme. In an additional analysis, we will adjust for the base-line score and the four stratification variables using an analysis of covariance. We will use independent t-test to compare the change scores between groups. We will also use the confidence interval approach as recommended for equivalence and non-inferiority trials[ 13 ] We will establish non-inferiority of interval exercise if the point estimate and its 95% confidence interval for difference between the change scores of the continuous and interval exercise group is smaller than the a priori determined boundary of clinical equivalence (see figure 3 ). If the 95% confidence intervals lie outside the boundaries of clinical equivalence we will establish clinical superiority of one exercise protocol. Figure 3 Illustration of the confidence interval approach to interpret results from randomised trials. The horizontal line indicates the difference between CRQ change scores between study groups. ± 0.5 points represent the predefined boundaries of equivalence. If the whole confidence interval is on the right of 0.5 points, interval exercise is not inferior to continuous exercise. If the whole confidence interval is within boundaries the two exercise protocols are of clinically equivalent effectiveness (upper three confidence intervals). Note that there can be a statistically significant difference between study groups but without any clinical relevance. Most methodologists and statisticians agree that the boundaries of equivalence should be defined as the minimal important difference and below the differences observed in previous trials comparing active to control treatments.[ 13 , 39 ] The minimal important difference is "the smallest difference in score in the outcome of interest that informed patients perceive as important, either beneficial or harmful, and which would lead the clinician to consider a change in the management"[ 40 ]. Therefore, the minimal important difference of our two main outcome measures for effectiveness have been established empirically and are around 0.5 for the CRQ domains scores [ 41 - 44 ] and 53 meters for the six-minute walk test[ 45 ]. A recent meta-analysis showed that respiratory rehabilitation with physical exercise leads to improvements of 50 meters in the six-minute walk test. Therefore we lower the boundaries for equivalence to ± 45 meters in order to have boundaries that are below the differences observed in previous trials comparing active to control treatments.[ 13 , 39 ] We will repeat the analysis with calculations of raw and adjusted differences for all other clinical and physiologic outcome measures and test for significant differences between groups using independent t-test if data are distributed normally. We will use both intention to treat and per protocol analysis to show equivalence in either case as recommended by others[ 13 , 39 , 46 , 47 ] Exercise tolerance Our null hypothesis is that patients equally experience high intensity continuous exercise and interval exercise as measured by the Exercise Tolerance Questionnaire. The alternative hypothesis is that interval exercise is associated with the experience of less limiting symptoms compared with high intensity continuous exercise. We will use independent t-tests to compare the measures for exercise tolerance (Exercise Tolerance Questionnaire, Borg scales for dyspnoea and leg pain) between groups. Again, we will assess the raw differences between groups in the primary analysis and adjust for baseline scores and the four stratification variables in the additional analysis. Confounder variables and effect modifiers Factors, which interfere with outcome measures, can distort the results if unevenly distributed between study groups. We use three approaches to control for confounders: First, we use randomisation to allocate patients. Second, we strengthen the randomisation using a computerised minimisation procedure with factors that are likely to influence the outcome measures (exercise capacity, pulmonary state and presence and absence of affective disorders). Third, we will collect a number of variables at baseline (age, gender, lung function, time since diagnosis, cumulative dose of oral steroids in previous three months, medication, cardiovascular, musculoskelettal and endocrine co-morbidities) that may modify the effect of exercise. We will compare the distribution of these potential effect modifiers between groups and statistically assess their influence on the outcome measures with multiple linear regression models. Sample size For calculating the required sample size, we use the formula for comparison of 2 means: n = [A + B] 2 * 2 * SD 2 /DIFF 2 , where n = the sample size required in each group (double for total sample), SD = standard deviation of the outcome variable, DIFF = size of desired difference between groups. A and B depend on the desired significance level and desired power, respectively. We base our sample size calculations on the CRQ. We used empirical data from our previous trial.[ 27 ], where we applied similar inclusion criteria for patients with COPD undergoing respiratory rehabilitation, to estimate variability of the CRQ (standard deviation of the CRQ domain scores between 0.8 and 1.2). A clinically sensible way to determine the size of desired difference between groups (DIFF in formula above) is based on the minimal important difference (0.5 on the scale from 1 to 7 for the CRQ)[ 41 - 43 , 48 ] A sample size of 44 patients in each group will allow showing a difference of 0.5 in CRQ scores between the groups, assuming a standard deviation of 0.8, with a power of 90% at a significance level of 5% (one-sided). Assuming a drop out rate of 15%.[ 27 ], the total minimal sample size increases to 104. This sample size will also allow detecting a difference of 45 meters in the six-minute walking test scores between the two treatment groups with a power of 95% at a significance level of 5% (one-sided) assuming a drop out rate of 15%. A priori sample size calculations usually provide only rough estimates. Therefore, we will recalculate sample size during the study when we have the data for 30 patients in each group. To this end, we will re-assess the standard deviation of the outcome variables and, if necessary, adjust the sample size accordingly (without breaking the randomisation code). Data collection and quality control We will implemented a series of measures to ensure high data quality. 1. Site investigators will collect the data using standardized forms. All data will be collected and entered into a single database in the Horten Centre by one investigator. A second investigator will validate completeness and accuracy of data extraction by checking 20% of extracted patient data. 2. Checklists with all the data to be collected will be provided for physical therapists and physicians. 3. Teaching sessions will be held regularly for physical therapists and physicians involved in data collection aiming at robust data collection. 4. Investigators meetings will be held monthly, or more frequently if needed, to discuss recruitment of patients, problems in conducting the study, acquisition of data, to check for consistency and completeness of data and for interim analyses. 5. E-mail will serve as the first line of non-urgent communication between research team members. 6. Monthly reports will be prepared by the principal investigator that will include the number of patients recruited, stage of follow up for each patient, notification regarding missing patient data and queries from data received. Discussion In the last 30 years, researchers made great efforts to study the effectiveness of respiratory rehabilitation compared to usual care. The meta-analyses of a recent systematic review[ 3 ] showed that respiratory rehabilitation with physical exercise leads to clinically significant improvements of HRQL as well as to significant improvements of functional and maximum exercise capacity. Research in respiratory rehabilitation should now focus on the evaluation of different exercise protocols. When an effective treatment such as respiratory rehabilitation is available, patients and clinicians are not confronted with the decision whether to start treatment or not, but with the decision on the most appropriate treatment. Therefore, clinicians need evidence from pragmatic randomised controlled trials directly comparing different exercise protocols at issue rather than evidence from (explanatory) trials comparing exercise with no exercise or usual care.[ 49 ] The proposed pragmatic trial will therefore provide important and needed guidance for decision-making in respiratory rehabilitation, in particular for those COPD patients who are severely impaired and initiate respiratory rehabilitation programmes. The evidence generated in the proposed clinical trial will also be relevant for the scientific community. From the 1970s up to the mid-1990s most investigators conducted explanatory clinical trials in order to better understand how and why physical exercise is effective in patients with COPD. After the recognition of its effectiveness the debate on the optimal exercise modality arose[ 6 ]. However, only few pragmatic trials have been conducted so far to advance the understanding of the relative benefit and downsides of different exercise protocols. With the proposed trial, we compare two clinically relevant interventions with the use of a clinical trial design that is useful for clinical decision making. Such trials are currently needed to gain consensus on the optimal exercise protocol. There is a need for randomised controlled trials to explore which exercise protocols are most effective for COPD patients. Another topic that has received little attention in respiratory rehabilitation trials so far is compliance to and subjective experience of physical exercise[ 9 ] Exercise protocols and adherence to it have not been described in detail in published studies[ 3 ] and therefore very little is known about the tolerance of different training modalities. This is despite the fact that physical exercise has long been considered unfeasible in patients with COPD. Beside the paucity of data on the effectiveness and tolerance of different exercise protocols, there is also a need for trials that are methodologically sound and rigorous. Most of the studies on exercise or respiratory rehabilitation in patients with COPD did not report on details of exercise tests and protocols. In addition, study design related issues that introduce bias such as description of the randomisation procedure, concealment of allocation, sample size calculations or blinding have not been addressed frequently. Therefore, we try to use strong epidemiological methods to minimize bias in our trial. Our study could make an important contribution to the understanding of physical exercise in patients with COPD and have a significant impact on the structure of respiratory rehabilitation and exercise programmes. Competing interests None declared. Abbreviations COPD = Chronic obstructive pulmonary disease HRQL = Health-related quality of life CRQ = Chronic Respiratory Questionnaire HADS = Hospital Anxiety Depression Score FT = Feeling Thermometer Authors' contributions MP drafted and revised the manuscript. All authors participated in development of research protocols and in the design of the study. MP, CZ and HS resolved statistical and methodological issues. All authors read and corrected draft versions of the manuscript and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC514611.xml
516774	The aminoguanidine carboxylate BVT.12777 activates ATP-sensitive K+ channels in the rat insulinoma cell line, CRI-G1	Background 3-guanidinopropionic acid derivatives reduce body weight in obese, diabetic mice. We have assessed whether one of these analogues, the aminoguanidine carboxylate BVT.12777, opens K ATP channels in rat insulinoma cells, by the same mechanism as leptin. Results BVT.12777 hyperpolarized CRI-G1 rat insulinoma cells by activation of K ATP channels. In contrast, BVT.12777 did not activate heterologously expressed pancreatic β-cell K ATP subunits directly. Although BVT.12777 stimulated phosphorylation of MAPK and STAT3, there was no effect on enzymes downstream of PI3K. Activation of K ATP in CRI-G1 cells by BVT.12777 was not dependent on MAPK or PI3K activity. Confocal imaging showed that BVT.12777 induced a re-organization of cellular actin. Furthermore, the activation of K ATP by BVT.12777 in CRI-G1 cells was demonstrated to be dependent on actin cytoskeletal dynamics, similar to that observed for leptin. Conclusions This study shows that BVT.12777, like leptin, activates K ATP channels in insulinoma cells. Unlike leptin, BVT.12777 activates K ATP channels in a PI3K-independent manner, but, like leptin, channel activation is dependent on actin cytoskeleton remodelling. Thus, BVT.12777 appears to act as a leptin mimetic, at least with respect to K ATP channel activation, and may bypass up-stream signalling components of the leptin pathway.	Background ATP-sensitive K + (K ATP ) channels are important regulators of cell function, coupling energy metabolism with electrical activity. K ATP channels are comprised of two proteins, derived from the sulphonylurea receptor (SUR) family and an inwardly rectifying K + channel (Kir6.x family), the exact composition of these being dependent upon tissue [ 1 , 2 ]. For example, pancreatic β-cells and insulin-secreting clonal cell lines express K ATP channels consisting of Kir6.2 and SUR1 subunits [ 3 ]. K ATP channels are present in numerous tissues and are the target for drugs that inhibit or increase channel activity [ 4 , 5 ]. The archetypal inhibitors of these channels are the sulphonylurea class of drugs, which bind to the SUR subunit of the channel. Modulation of K ATP channel activity in pancreatic β-cells has profound effects on insulin secretion and glucose homeostasis [ 6 ]. Sulphonylureas such as tolbutamide and glibenclamide inhibit channel activity, resulting in β-cell depolarization, increased electrical activity, enhanced calcium entry and consequently increased insulin secretion [ 7 ]. In contrast, pancreatic β-cell K ATP channel activation induces hyperglycaemia in animals and man [ 8 ]. This latter action is caused by membrane hyperpolarization, reduction in cell excitability and decreased intracellular calcium resulting in reduced secretion of insulin. Such effects have been reported following application of the benzothiadiazine, diazoxide, which has been used on occasion to treat persistent hyperinsulinemic hypoglycaemia of infancy [ 8 ]. It has been demonstrated that diazoxide interacts with the sulphonylurea receptor subunit, SUR1, encompassing transmembrane domains 6–11 and the first nucleotide binding fold [ 9 ]. A similar conclusion has also been reached using a novel diazoxide analogue [ 10 ]. The presence of K ATP channels in many other tissues, notably muscle and central neurons, has stimulated interest in the development of novel, selective K ATP channel openers for the treatment of various diseases [ 10 , 11 ]. The ob gene product leptin has been demonstrated to activate K ATP channels in pancreatic β-cells [ 12 ] and insulin-secreting cell lines [ 13 ], consistent with a potential role in modifying insulin secretion [ 14 ]. One of the primary functions for this hormone is its role in the regulation of food intake and body weight [ 15 ]. Interestingly, leptin also activates K ATP channels of hypothalamic glucose-responsive neurones [ 16 , 17 ] indicating a possible role for this channel in the control of energy homeostasis and body weight. In addition, Kir6.2 knock-out mice have deficits in central glucose sensing leading to loss of glucose mediated feeding response and a defective hypoglycaemic compensatory response [ 18 ]. These latter findings suggest that hypothalamic K ATP channels may also be an important target for drug manipulation with respect to centrally driven control of glucose and energy homeostasis. The aminoguanidine carboxylate, BVT.12777 (Figure 1 ), is one of a series of structurally related molecules based on the anti-diabetic/anti-obesity agent 3-guanidinopropionic acid [ 19 ], which, like leptin, have been demonstrated to reduce body weight in obese diabetic ( ob/ob ) mice [ 20 ]. Here we demonstrate that BVT.12777 opens K ATP channels in the CRI-G1 insulin secreting cell line, a useful model for pancreatic β-cells [ 21 ], and for analysing the mechanism by which leptin opens K ATP channels [ 13 , 22 , 23 ]. Figure 1 Structure of BVT.12777 ([2-(hydrazinoiminomethyl)-hydrazino] acetic acid) Results BVT.12777 activates K ATP channels Under current clamp conditions with 5 mM ATP in the pipette solution to maintain K ATP channels in the closed state, the mean resting potential was -38.7 ± 1.7 mV (n = 10), similar to values reported in previous studies [ 13 , 22 ] under these recording conditions. Application of BVT.12777 (100 μM) hyperpolarized CRI-G1 cells (Figure 2A ) to -66.3 ± 2.7 mV (n = 10). Examination of the voltage-clamped macroscopic currents indicates that prior to the addition of BVT.12777 the slope conductance of the cells was 0.43 ± 0.03 nS (n = 10), and following exposure to BVT.12777 (100 μM), this increased to 3.45 ± 1.17 nS (n = 10). The reversal potential (obtained from the point of intersection of the current-voltage relationship) associated with the BVT.12777-induced conductance increase (Figure 2A ) was -78.5 ± 0.8 mV (n = 10), close to the calculated value for E k of -84 mV in this system, indicating increased K + conductance. CRI-G1 cells responded to BVT.12777 in an all or none manner, with cells undergoing full hyperpolarization and increase in conductance, at all concentrations (100 – 300 μM) examined. Such an effect has also been reported for leptin on CRI-G1 cells [ 13 ]. Removal of BVT.12777 from the bath solution did not fully recover the membrane potential and conductance to control values over the next 15–30 minutes (not shown). Application of the K ATP channel inhibitor, tolbutamide (100 μM) during BVT.12777 exposure (Figure 2A ) completely reversed the BVT.12777-induced hyperpolarization and decreased conductance, to -41.0 ± 4.8 mV (n = 5) and 0.58 ± 0.07 nS (n = 5) respectively, values indistinguishable from control (P > 0.05). These data indicate that BVT.12777 increases K ATP current in this cell line. This is demonstrated more clearly in cell-attached recordings from CRI-G1 cells, where bath application of BVT.12777 (100 μM) resulted in activation of single K ATP channel currents (Figure 2B ; n = 7). The increase in channel activity was evident within 5 minutes of drug application, was sustained over the time course of exposure (~30 minutes) and was not immediately reversed following removal of the drug. Figure 2C shows mean channel activity (N f .P o ), normalised to the control for each recording, plotted against time of exposure to BVT.12777. BVT.12777 activation of K ATP channels was demonstrated to be reversibly inhibited by 100 μM tolbutamide (n = 4; Figure 2B,2C ). Identical control experiments, in the absence of BVT.12777, resulted in no significant effect on K ATP channel activity, over a 30-minute test period (n = 8; P > 0.05). Figure 2 BVT.12777 activates a tolbutamide-sensitive K+ current A , the upper trace shows a current clamp recording of a CRI-G1 cell following dialysis with a 5 mM ATP-containing solution. In this and subsequent current clamp figures the trace begins approximately 5 min after formation of the whole-cell configuration. Application of BVT.12777 (100 μM) for the time indicated hyperpolarized the cell from -50 mV to -76 mV, an action readily reversed by tolbutamide (100 μM), which returned membrane potential to -54 mV. Washout of all drugs from the bath resulted in a membrane potential of -70 mV, indicating the lack of reversibility of BVT.12777. The lower plot is the current-voltage relationship for the voltage clamped currents. Cells were voltage clamped at -50 mV and 10 mV steps of 100 ms duration were applied every 200 ms (range -120 to -30 mV). BVT.12777 increased the membrane conductance relative to control and tolbutamide reversed this BVT.12777-induced conductance increase with a reversal potential of -78 mV. B , cell-attached recording from a CRI-G1 cell, at 10 mV applied to the recording pipette. Single channel openings are shown as downward deflections. Addition of 100 μM BVT.12777 induced an increase in channel activity (N f .P o ) from 0.17 in control to 0.31, and 1.25 at 10 and 20 minutes respectively, after BVT addition. Application of 100 μM tolbutamide induced a substantial inhibition of activity (to 0.02), which was reversed on washout of all drugs, with activity increasing to 0.74. The symbol C refers to the closed state of the channel in this and subsequent figures. C , diary plot of N f .P o against time from cell-attached experiments in the presence and absence of BVT.12777, where channel activity was calculated every 2 minutes. Each point is the mean of 4–7 separate determinations. The effect of BVT.12777 on K ATP channel activity in excised membrane patches was also examined. Recordings were made from inside-out patches in symmetrical (140 mM KCl in pipette and bath solutions) K + at a membrane potential of -40 mV. K ATP channels were identified by inhibition of channel activity following application of 100 μM MgATP to the inner membrane aspect of the patch, which reduced normalised N f P o from 1.0 to 0.23 ± 0.05 (n = 4; P < 0.05). Subsequent application of 100 μM BVT.12777, in the continued presence of MgATP, induced a gradual increase in K ATP channel activity (Figure 3 ), to levels similar to that of control (in the absence of MgATP). For example 15 minutes after 100 μM BVT.12777 application normalised mean channel activity had recovered to 1.18 ± 0.46 (n = 4). In experiments where no drug was added, K ATP channel currents, in the presence of 100 μM MgATP, did not activate spontaneously (n = 4). Figure 3 BVT.12777 activates K ATP channels in inside-out patches Continuous single channel currents recorded from an inside-out patch at a holding potential of -40 mV. Application of 100 μM MgATP reversibly inhibited channel activity by >90%, demonstrating K ATP identity. Addition of 100 μM BVT.12777, in the presence of 100 μM MgATP to the cytoplasmic aspect of the patch resulted in K ATP channel activation. N f .P o values were 2.96 (control, after first MgATP challenge), and 0.25 in the presence of MgATP, which increased to 0.72, 1.06 and 2.74 at 5, 10 and 20 minutes respectively, after BVT.12777 addition. BVT.12777 activates K ATP channels independently of PI 3-kinase activity Leptin and diazoxide hyperpolarized CRI-G1 cells, in a manner similar to that of BVT.12777 (data not shown). Leptin (10 nM) induced a hyperpolarization from a mean membrane potential of -47.6 ± 1.6 mV to -68.5 ± 1.9 mV (n = 8; P < 0.05), and application of tolbutamide (100 μM) reversed this action, returning the membrane potential to -47.5 ± 1.9 mV (n = 4). Diazoxide (200 μM) rapidly hyperpolarized CRI-G1 cells from a mean membrane potential of -49.9 ± 1.7 mV to -74.0 ± 1.5 mV (n = 6; P < 0.05), with tolbutamide (100 μM) also reversing this action, returning membrane potential to -46.9 ± 3.8 mV (n = 6). Leptin, but not diazoxide activation of CRI-G1 K ATP channels is PI3K dependent [ 22 , 23 ]. Thus, we investigated whether BVT.12777 activates K ATP channels in CRI-G1 cells by direct (like diazoxide) or indirect (like leptin) mechanisms. Pre-incubation of CRI-G1 cells (20 min) with inhibitors of PI 3-kinase, wortmannin (10 nM) or LY294002 (10 μM) had no significant effect on the mean resting membrane potential or slope conductance of CRI-G1 cells and did not prevent BVT.12777 from causing hyperpolarization and increased cell conductance (Figure 4A ). In the presence of 10 nM wortmannin, values for mean membrane potential and slope conductance were -44.3 ± 1.2 mV (n = 6) and 0.86 ± 0.10 nS (n = 5), and addition of 200 μM BVT.12777 hyperpolarized cells to -68.9 ± 0.8 mV (n = 6) with an increase in slope conductance to 3.10 ± 0.38 nS (n = 5). Identical results were obtained in the presence of 10 μM LY294002 (data not shown), with corresponding control values of -40.8 ± 2.8 mV (n = 6) and 0.79 ± 0.11 nS (n = 4), and in the presence of 200 μM BVT.12777, -67.9 ± 0.6 mV (n = 6) and 2.69 ± 0.35 nS (n = 4) for membrane potential and slope conductance respectively. In all experiments (i.e with either PI3K inhibitor) addition of tolbutamide (100 μM) recovered the membrane potential (-41.8 ± 1.5 and -34.0 ± 1.7 mV; n = 6) and slope conductance (0.89 ± 0.11 (n = 5) and 0.58 ± 0.06 (n = 4) nS) for wortmannin and LY294002 respectively, to values indistinguishable from controls (P > 0.1). Cell-attached recordings from CRI-G1 cells also show that wortmannin (10 – 100 nM) did not occlude BVT.12777 activation of K ATP channels (Figure 4B ). Mean channel activity in the presence of wortmannin (10 nM) was 0.02 ± 0.00 which increased to 0.16 ± 0.02, 20 minutes after exposure to 100 μM BVT.12777 (n = 3; P < 0.05). Control experiments where no BVT.12777 was added show no change in channel activity over a 30-minute period (N f .P o = 0.01 ± 0.00 and 0.04 ± 0.00 after 5 and 30 minutes respectively; n = 4). Figure 4 Wortmannin does not inhibit BVT.12777 activation of K ATP A , current clamp record of a CRI-G1 cell dialysed with 5 mM MgATP, following exposure of cells to 10 nM wortmannin for 15–20 minutes. Application of BVT.12777 (200 μM), in the continued presence of wortmannin hyperpolarized the cell from -46 to -77 mV. Tolbutamide (100 μM), applied after the BVT-induced hyperpolarization, recovered the membrane potential (to -40 mV). B , cell-attached recordings from CRI-G1 cells, following exposure of cells to 10 nM wortmannin for 15–20 minutes. Upper trace; in the continued presence of wortmannin, N f .P o was 0.01 and 0.03 after 5 and 30 minutes respectively. Lower trace, application of BVT.12777 (100 μM) to cell-attached recording in the presence of 10 nM wortmannin resulted in K ATP activation, with N f .P o values of 0.01, 0.12 and 0.27 prior to, and 10 and 30 minutes after, BVT.12777, respectively. Addition of 100 nM wortmannin did not inhibit channel activity. Heterologously expressed K ATP currents are not activated by BVT.12777 Oocytes injected with Kir6.2 and SUR1 cRNAs were challenged with sodium azide (3 mM) to elicit a reversible increase in current, which was completely blocked by 1 μM glibenclamide or 0.5 mM tolbutamide, indicating that the current was due to K ATP activation, as described previously [ 24 , 25 ]. In oocytes, previously exposed to sodium azide in order to verify Kir6.2-SUR1 expression, application of BVT.12777 (10 μM – 1 mM) did not produce any consistent increase in K ATP current (n = 16; data not shown). Consequently, we utilized an alternative expression system, the HEK 293 cell line [ 25 ]. Application of BVT.12777 (100 μM) to the bathing solution using the cell-attached recording configuration resulted in no significant increase in mean channel activity above control levels over a 30-minute period, although subsequent addition of sodium azide (3 mM) did cause a rapid increase in channel activity, which was reversed by the addition of 100 μM tolbutamide (n = 4, data not shown). Similarly, application of BVT.12777 in the presence of 0.1 mM MgATP to inside-out patches from HEK 293 cells transiently expressing Kir6.2-SUR1, did not cause activation of channel activity following 30 minutes exposure (n = 4; data not shown). Thus BVT.12777 does not appear to be capable of activating heterologously expressed Kir6.2-SUR1 currents. MAPK does not mediate BVT.12777-activation of K ATP Exposure of CRI-G1 cells to BVT.12777 (100 μM) for up to 30 minutes had no consistent effect on the phosphorylation of enzymes downstream of PI3K (PKB and its downstream target, GSK3), but did increase the phosphorylation of STAT3 (n = 4) and MAPK (n = 4; data not shown). These data are in agreement with the lack of BVT.12777 sensitivity to PI3K inhibitors on activation of K ATP channels. However, activation of MAPK has been implicated as a significant intermediate for both insulin and leptin signalling pathways in various cell types [ 26 - 29 ]. Thus, we examined the effect of UO126, a potent and specific inhibitor of the activation of the classical MAPK cascade [ 30 ], on BVT.12777 opening of K ATP channels. Application of UO126 (25 μM) inhibited approximately 90 % of K ATP channel activity in cell-attached or inside-out recordings, whereas 1–10 μM UO126, concentrations that suppresses activation of MAPKK [ 30 ], had no significant effect on channel activity (data not shown). Control cell attached recordings had a mean channel activity of 0.07 ± 0.02, which increased to 1.29 ± 0.82 (n = 3) in the presence of BVT.12777 (100 μM). Subsequent application of UO126 (1 μM) in the continued presence of BVT.12777 did not alter channel activity (data not shown), over a 15-minute period (N f .P o was 1.86 ± 1.45 (n = 3) and 2.44 ± 2.00 (n = 3), at 5 and 15 minutes respectively; P < 0.05). In addition, increasing UO126 to 10 μM had no effect on BVT.12777 induced K ATP channel activation. BVT.12777 activation of K ATP channels is dependent on actin cytoskeleton dynamics Leptin activation of K ATP channels in the CRI-G1 cell line is dependent upon reorganisation of the cytoskeleton, a process downstream from PI3K activation [ 31 ]. Therefore, we examined whether BVT.12777 opening of CRI-G1 K ATP channels occurs through alteration of actin filament dynamics. For this series of experiments the heptapeptide mushroom toxin phalloidin [ 32 ] was used to stabilise the polymerised form of actin (F-actin). As phalloidin is membrane-impermeant, it was directly applied to the internal aspect of the cell membrane. In whole-cell experiments, 10 μM phalloidin was added to the electrode solution and allowed to dialyse into the cell. The mean resting potential and slope conductance were -38.0 ± 0.6 mV and 0.66 ± 0.04 nS (n = 4) respectively, and following addition of 200 μM BVT.12777 no significant change in these parameters was observed (Figure 5A ), with a mean membrane potential of -41.7 ± 1.1 mV and slope conductance of 0.60 ± 0.08 nS (n = 4; P > 0.05). The presence of phalloidin (10 μM) in the bath solution also prevented K ATP channel activation by BVT.12777 in the inside-out isolated patch configuration (Figure 5B ). Application of 0.1 mM MgATP to the cytoplasmic aspect of inside-out patches caused 97.5 ± 2.1% inhibition of K ATP channel activity (n = 3; P < 0.05) and subsequent addition of 10 μM phalloidin had no further effect, as reported previously [ 29 ]. Subsequent addition of BVT.12777 (100 μM) failed to increase K ATP channel activity, with mean Nf.Po values of 0.06 ± 0.05 and 0.03 ± 0.01 in the absence and presence of BVT.12777 respectively (n = 3; P > 0.05). In contrast, the direct K ATP channel opener, diazoxide activates K ATP channels in the presence of phalloidin. In whole-cell experiments (Figure 5C ), diazoxide (200 μM) hyperpolarized CRI-G1 cells from a mean membrane potential of -42.6 ± 0.1 mV to -70.1 ± 0.8 mV (n = 4; P < 0.05), and increased slope conductance from 0.87 ± 0.23 to 7.39 ± 0.72, actions reversed by tolbutamide (100 μM). Figure 5 Phalloidin prevents BVT.12777 activation of K ATP A , current clamp record of a CRI-G1 cell dialysed with 5 mM MgATP and 10 μM phalloidin. Application of BVT.12777 (200 μM) had no effect on the membrane potential of the cell (-41 mV) in the presence of phalloidin. B , continuous single channel currents recorded from an inside-out patch at a holding potential of -40 mV. Application of 100 μM MgATP reversibly inhibited N f .P o from 1.25 to 0.02. Addition of 10 μM phalloidin and subsequently 100 μM BVT.12777, in the presence of 100 μM MgATP, to the cytoplasmic aspect of the patch resulted in no effect on K ATP , with N f .P o values of 0.01 and 0.03 respectively. C , current clamp record of a CRI-G1 cell dialysed with 5 mM MgATP and 10 μM phalloidin. Application of diazoxide (200 μM) induced rapid cell membrane hyperpolarization, from -55 to -72 mV, an action reversed (to -45 mV) by tolbutamide (100 μM). F-actin is disrupted by BVT.12777 The prevention of BVT.12777-induced K ATP activation by phalloidin mirrors the effect of this toxin on leptin activation of K ATP [ 31 ]. Thus, we visualised F-actin by staining with rhodamine-conjugated phalloidin. In untreated CRI-G1 cells there was pronounced phalloidin-positive labelling of the cell membrane, with more diffuse, granular staining within the cytoplasm (Figure 6A ). In contrast, cells treated with BVT.12777 (100 μM) or leptin (10 nM) for 40 min showed a marked reduction in phalloidin fluorescence intensity, with disjointed labelling at the cell membrane (Figure 6A ). The actin filament disrupter cytochalasin B [ 33 ] also reduced the intensity of phalloidin labelling but in a more punctate manner on visualisation of treated cells compared with controls (data not shown). Analysis of the mean fluorescence intensity at the cell membrane following the actions of BVT.12777 and leptin demonstrated that both treatments caused a significant reduction of the intensity of rhodamine-phalloidin labelling, by 43.0 ± 4.2% (n = 6; P < 0.05) and 62.2 ± 6.0% (n = 6; P < 0.05), respectively, compared to untreated cells (Figure 6B ). However, the directly acting K ATP channel opener, diazoxide did not cause disruption of the actin cytoskeleton (Figure 6A,6B ), with a relative intensity of rhodamine-phalloidin staining of 0.98 ± 0.16 (P > 0.05). Figure 6 BVT.12777 disrupts the actin cytoskeleton A , images of rhodamine-conjugated phalloidin fluorescence in CRI-G1 cells in control conditions and following incubation with leptin (10 nM), BVT.12777 (100 μM) or diazoxide (200 μM) for 30 minutes. All panels show representative X-Y images. Note the marked reduction in phalloidin staining in cells pre-treated with leptin or BVT.12777, and not diazoxide. Scale bars are 50 μm. B , histogram comparing the normalised fluorescence intensity relative to control in the membrane periphery of randomly selected CRI-G1 cells for each condition; (control (n = 13; cells = 195), 10 nM leptin (n = 6; cells = 90), BVT.12777 (n = 6; cells = 90) and 200 μM diazoxide (n = 4, cells = 60). Error bars indicate s.e.m. and * significance of P < 0.001. Discussion BVT.12777 induced hyperpolarization of CRI-G1 cells, with an associated increase in K + conductance, an action likely caused by the activation of K ATP channels, as the sulphonylurea tolbutamide completely reversed its effects. Cell-attached and inside-out single channel current recordings demonstrate directly that BVT.12777 activates K ATP channels. The increased K ATP current generated in isolated membrane patches resembles the effects of K ATP activators such as diazoxide [ 34 ] and sodium azide [ 35 ], which have also been shown to activate insulinoma or pancreatic β-cell K ATP channels in isolated patches in the presence of Mg-ATP. Thus, although not tested here, BVT.12777 as an activator of K ATP would be expected, as observed for diazoxide, to inhibit insulin release from CRI-G1 cells stimulated by metabolizable substrates or tolbutamide [ 36 ], although this would clearly be dependent on its action on other β-cell conductances, notably calcium channels. BVT.12777 activation of K ATP channels was only slowly reversed on withdrawal of the drug, unlike the actions of diazoxide or sodium azide, which are rapidly reversed on washout [ 35 , 36 ]. Indeed, following removal of BVT.12777 in the absence or presence of tolbutamide, enhanced K ATP channel activity was apparent for a considerable time. The slow reversibility on washout of BVT.12777 resembles the effects of the hormone leptin on CRI-G1 cell membrane potential and K ATP channel activation [ 13 ]. Leptin, via activation of the main signalling form of the leptin receptor (ObRb), has been shown to increase the phosphorylation of STAT3, MAPK and to stimulate PI3K pathways in various peripheral tissues, cell lines [ 37 ], and in hypothalamic neurones [ 38 ]. BVT.12777 although stimulating phosphorylation of STAT3 and MAPK did not stimulate PI3K dependent pathways as demonstrated by the lack of effect on the phosphorylation status of the PI3K output indicators, PKB and GSK3. It is unclear at present how this molecule induces STAT3 and MAPK phosphorylation. As K ATP activation by BVT.12777 is rapid and occurs in isolated membrane patches it is unlikely that any JAK-STAT pathway (which drives changes in transcription) contributes to this action. Leptin activation of K ATP channel currents in CRI-G1 cells has previously been shown to be independent of MAPK, but prevented by the inhibitors of PI3K [ 22 ]. However, BVT.12777 activation was not only insensitive to inhibition by the MAPKK inhibitor, UO126, it was also insensitive to the presence of the PI3-kinase inhibitors, wortmannin and LY294002, at concentrations sufficient to prevent leptin activation of K ATP in this cell line. These data led us to suspect that BVT.12777, irrespective of its ability to initiate various signalling cascades in this cell line, increased K ATP channel activity by a more direct effect on the channel subunits in a manner analogous to diazoxide, which is purported to interact directly with the SUR1 subunit [ 9 , 10 ]. This possibility was tested by heterologous expression of the β-cell subunits of K ATP channels, Kir6.2 and SUR1, in Xenopus oocytes, a commonly utilised expression system for electrophysiological studies of these recombinant channels [ 24 , 25 ]. However, BVT.12777 did not activate Kir6.2-SUR1 currents in oocytes, demonstrated to express functional K ATP channel currents. Thus we explored this question further by utilising a second heterologous expression system for Kir6.2-SUR1, HEK293 cells. Recordings from inside-out patches demonstrated that BVT.12777 did not activate Kir6.2-SUR1 currents in the presence of Mg-ATP, in contrast to diazoxide [ 39 ] or sodium azide [ 35 ]. Overall these data strongly suggest that expression of the K ATP channel subunits, Kir6.2 and SUR1 are insufficient per se to bring about sensitivity to BVT.12777, and indicate that this opener may activate this channel type by an indirect mechanism (which is not available in oocytes or HEK cells). Although the activation of K ATP channels by leptin in CRI-G1 cells is PI3-kinase dependent the lipid products of this enzyme system, such as PtdIns(3,4,5)P 3 also do not interact directly with K ATP channels [ 22 ]. Recent studies demonstrate that both leptin and PtdIns(3,4,5)P 3 increase K ATP channel activity indirectly, through changes in cytoskeletal dynamics [ 31 ]. It is well established that many ion channels and transporters are anchored in the membrane by either direct or indirect association with the cytoskeleton. In addition, there is growing evidence that altering the integrity of cytoskeletal elements, in particular actin filaments, can modulate the activity of a variety of ion channels [ 40 ] and receptors [ 41 ]. For example, disruption of actin filaments with cytochalasin is shown to increase K ATP channel activity in cardiac myocytes [ 42 ] and CRI-G1 cells [ 31 ]. Indeed, a number of lipid kinases, including PI 3-kinase, are also localised to the cytoskeleton and their activities are modulated by a variety of cytoskeletal proteins, especially those associated with actin [ 40 ]. Actin filament structure is controlled by reversible polymerisation of G-actin, which forms F-actin, and this process is under the dynamic control of various actin-binding proteins [ 43 ]. The heptapeptide mushroom toxin phalloidin [ 32 ] binds to filamentous F-actin with high affinity and stabilises the actin in this form. The addition of phalloidin to the intracellular aspect of CRI-G1 cells prevented BVT.12777, but not diazoxide, from activation of K ATP channel currents in whole cell and inside out recording configurations indicating that this molecule likely causes the opening of K ATP channels by a membrane delimited alteration of cytoskeletal dynamics. This mechanism of action is identical to that proposed for leptin and PtdIns(3,4,5)P 3 activation of K ATP in this cell line [ 31 ]. Fluorescence staining of CRI-G1 cells with rhodamine-conjugated phalloidin revealed disassembly of actin filaments by both BVT.12777 and leptin, but not diazoxide. These data provide direct support for an important role for cytoskeletal dynamics in the control of K ATP channel activity by both leptin and BVT.12777. The lack of effect of diazoxide on the actin filament structure is also supportive of this opener acting directly on the K ATP channel subunits. Conclusions BVT.12777 activation of K ATP channels in CRI-G1 cells was evident regardless of whether it was applied to the external or internal surface of the cell. BVT.12777 signalling to K ATP channels is not mediated by PI 3-kinase or MAPK, but does appear to depend on actin filament re-modelling. As leptin hyperpolarizes a sub-population of hypothalamic neurones by opening K ATP channels [ 16 ], it is feasible that at least part of the anti-obesity action of BVT.12777 may be through the activation of this potassium channel. Furthermore, as BVT.12777 acts downstream of PI3K, such an agent may act to overcome the putative central leptin resistance associated with the obese state [ 37 ]. Thus, although BVT.12777 and its close structural analogues are unlikely per se to be useful anti-obesity agents as they display hepatotoxicity [ 44 ], understanding the general principles underlying their mechanism of action may reveal clues for future anti-obesity drug development. Methods Cell culture and transfection Cells from the insulin secreting cell line, CRI-G1, and the human embryonic kidney cell line, HEK 293, were grown as described previously [ 25 , 35 ]. The preparation of mouse Kir6.2 (provided by Professor F. Ashcroft, University of Oxford), rat SUR 1 (provided by Dr G. Bell, University of Chicago) and CD4 cDNAs and transfection procedures were as described by [ 25 ]. Transfected cells were selected by visible binding of anti-CD4 coated beads (Dynal, Oslo) following incubation with the beads for 20 min. Oocyte collection and preparation Ovarian lobes were removed from mature female Xenopus laevis frogs (Blades Biological, UK) following killing of the animal by destruction of the brain. The use of animals was in accordance with the Home Office Animals (Scientific Procedures) Act (1986) and approved by the local ethics committee. Separation and selection of oocytes and the preparation and injection of cRNAs were performed as described by [ 25 ]. Western blotting CRI-G1 cells, in normal saline (containing in mM; NaCl 135, KCl 5, MgCl 2 1, CaCl 2 , 1, HEPES 10 with glucose 10 (pH 7.4) were treated with BVT.12777 (100 μM) for 0, 1, 5, 15 or 30 minutes and whole-cell extracts were prepared as described [ 23 ]. Proteins (10 μg) were suspended in loading buffer (Invitrogen) and after denaturation, loaded on to NuPage 4–12% Bis-Tris mini-gels (Invitrogen) and run at 200 V for 1 hr. Subsequently, proteins were transferred to Hybond-C Extra nitrocellulose membranes (Amersham) at 25 V for 80 minutes at room temperature. Membranes were incubated in blocking buffer (5% non-fat milk in TBST (20 mM Tris HCl, 150 mM NaCl, 0.5% Tween, pH 7.4)) for 1 hr at room temperature after which antibodies to phospho-MAPK, phospho-STAT3, phospho PKB, phospho-GSK3 and PKB (all at 1:1000) were applied at 4°C with gentle shaking, overnight. The membranes were washed with TBST (4 × 30 minutes) and incubated for 1 hr at room temperature with HRP conjugated ImmunoPure goat anti-rabbit IgG (1:5000). After washing with TBST (5 × 15 minutes), immunoreactive bands were visualised by the enhanced chemiluminescence (ECL) detection reagent (Amersham). Cytoskeletal fluorescence imaging and analysis CRI-G1 cells were gently washed in normal saline (containing in mM): NaCl 135, KCl 5, MgCl 2 1, CaCl 2 1, HEPES 10, pH 7.4, and incubated for 40 min with either 100 μM BVT.12777, 10 nM leptin, 200 μM diazoxide or 3 mM sodium azide for 30 min with the cytoskeletal disrupter, cytochalasin B (10 μM). Cells were then fixed, permeabilised, stained with rhodamine-conjugated phalloidin (2.66 U ml -1 ) and visualised using a BioRad Microradiance, confocal imaging system as described by [ 31 ]. The intensity of rhodamine-conjugated phalloidin staining in the plasma membrane was determined using BioRad Lasersharp processing software (Bio-Rad, CA, USA). Analysis lines were drawn along randomly selected regions of the plasma membrane and the fluorescence intensity determined. A histogram giving the mean fluorescence intensity was constructed for a minimum of 5 cells on each stimulated or control dish on at least 3 separate occasions. Within a given experimental series all conditions for capturing images were constant. In order to allow for quantification of experimental data obtained on separate days, the results were normalised relative to the mean plasma membrane fluorescence measured in the control cells for each day and presented as mean ± S.E.M. Statistical analyses were performed using Student's unpaired t test. p < 0.05 was considered significant. Electrophysiological recording and analysis Whole cell currents from Xenopus oocytes were measured using a two-electrode voltage clamp technique as described by [ 25 ]. Recordings were made in a high-potassium bath solution, KD96 containing (mM): KCl 96, NaCl 2, CaCl 2 1.8, HEPES 5 (pH 7.4 with KOH). Working concentrations of drugs were prepared in KD96 and superfused into the bath. Whole-cell current-clamp recordings with excursions to voltage clamp mode were used to monitor membrane potential and macroscopic currents from CRI-G1 cells. Cell-attached and excised inside-out recordings were made from CRI-G1 cells and HEK cells expressing Kir6.2 and SUR1 to examine single channel responses as described previously [ 25 , 35 ]. Single channel data were analysed for current amplitude and channel activity (N f .P o ; where N f is the number of functional channels in the patch and P o is the open probability) as described previously [ 45 ]. All data were normalised to control and are expressed as mean ± S.E.M. Statistical analyses were performed using Student's unpaired t test. P < 0.05 was considered significant. Recording electrodes were pulled from borosilicate glass and had resistances of 2–5 MΩ for whole cell recordings and 7–10 MΩ for cell-attached and inside-out experiments when filled with electrolyte solution. The pipette solution for whole-cell recordings comprised (in mM): KCl 140, MgCl 2 0.6, CaCl 2 2.73, Mg-ATP 5.0, EGTA 10, HEPES 10, pH 7.2 (free [Ca 2+ ] of 100 nM), whereas for single channel recordings the pipette solution contained (in mM): KCl 140, CaCl 2 1, MgCl 2 1, HEPES 10, pH 7.2. The bath solution for whole-cell and cell-attached recordings was normal saline whereas for inside-out patches the bath solution contained (in mM): KCl 140, MgCl 2 1, CaCl 2 2, EGTA 10, HEPES 10, pH 7.2 (free [Ca 2+ ] of 30 nM). All solution changes were achieved by superfusing the bath with a gravity feed system at a rate of 10 ml min -1 , which allowed complete exchange within 2 min. All experiments were performed at room temperature (22–25°C). Antibodies & drugs Anti-PKB, which recognises all three isoforms of PKB, and the phospho-specific PKB (Thr308), GSK3α/β (Ser21/9), STAT3 (Tyr705) and p44/42 MAPK (Thr202/Tyr204) antibodies were obtained from Cell Signalling Technology Inc. Recombinant human leptin, wortmannin and LY 294002 were obtained from Novachem-Calbiochem and BVT.12777 ([2-(hydrazinoiminomethyl) hydrazino] acetic acid) was a gift from Biovitrum (Stockholm, Sweden). Tolbutamide, Mg-ATP, diazoxide, sodium azide, phalloidin and cytochalasin B were obtained from Sigma. Rhodamine-conjugated phalloidin was obtained from Molecular Probes and UO126 from Promega. BVT.12777 was prepared as a 100 mM stock solution in normal saline and stored at -70°C prior to use. Leptin was prepared as a 10 μM stock solution in normal saline containing 0.2 % bovine serum albumin as carrier. Rhodamine-conjugated phalloidin (200 U ml -1 ) and LY 294002 (10 mM) were stored as stock solutions in 1% methanol at -20°C. Cytochalasin B was stored as a 10 mM stock solution, and diazoxide and tolbutamide as 100 mM solutions, all in DMSO at 2–4°C. Mg-ATP was stored at -20°C as a 100 mM solution in 10 mM HEPES (pH 7.2). Wortmannin and UO126 were stored as 10 mM stock solutions in Me 2 SO at -20°C. List of abbreviations used CRI-G1, Cambridge Rat Insulinoma-G1; GSK3, glycogen synthase kinase-3; HEK293, human embryonic kidney 293; JAK, janus kinase; K ATP , ATP-sensitive potassium; Kir6.2, potassium channel inward rectifier-6.2; MAPK, p42, p44 mitogen-activated protein kinase; MAPKK, MAPK kinase; ObRb, Obese (leptin) receptor-b; PKB, protein kinase B; PI3K, phosphatidylinositol 3-kinase; PtdIns(3,4,5)P 3 , phosphatidylinositol 3,4,5 tris-phosphate; STAT3, signal transducer and activator of transcription-3; SUR, sulphonylurea receptor Authors' contributions JK carried out the majority of the electrophysiology and cytoskeletal fluorescence studies. HL carried out the western blot experiments. TT and JH participated in the electrophysiological experiments. CS participated in the design and implementation of the western blot experiments. MA conceived of the study, participated in its design and co-ordination and drafted the manuscript. All authors read and approved the final manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC516774.xml
554976	Comparing strategies for United States veterans' mortality ascertainment	Background We aimed to determine optimal strategies for complete mortality ascertainment comparing death certificates and United States (US) Veterans Administration (VA) records. Methods We constructed a cohort of California veterans who died in fiscal year (FY) 2000 and used VA services the year before death. We determined decedent status using California death certificates linked to VA utilization data and the VA Beneficiary Identification and Records Locator System (BIRLS) death file. We compared the characteristics of decedents who would not have been identified by either single source (e.g., VA BIRLS alone or California death certificates alone) with the rest of the cohort. Results A total of 8,813 veteran decedents were identified from both VA decedent files and death certificates. Of all decedents, 5,698 / 8,813 (65%) veterans were identified in both source files, but 2,426 / 8,813 (28%) decedents were not identified in VA BIRLS, and 689 / 8,813 (8%) were not identified in death certificates. Compared to the rest of the cohort, decedents whose mortality status was ascertained through either single source differed by race / ethnicity, marital status, and California residence. Clinically, veterans identified from either single source had less comorbidity and were less likely to have been users of VA inpatient or long term care, but equally or more likely to have been users of VA outpatient services. Conclusion As single sources, VA decedent files and death certificates each provided an incomplete record, and death ascertainment was improved by using both source files. Potential bias may vary depending on analytic interest.	Introduction Clinicians, healthcare administrators, researchers, regulators and policymakers are concerned with optimizing mortality ascertainment using administrative data. In addition to its clinical importance, mortality informs program planning, quality assessment and improvement, and public reporting [ 1 - 8 ]. Veterans are an important, vulnerable population in which mortality has been examined as a function of race / ethnicity, service characteristics, access, and quality of care. Valid, complete reporting is critical to the success of such endeavors, and limitations in using death certificates have been acknowledged [ 9 , 10 ], although VA mortality data is generally regarded as accurate [ 11 - 15 ]. To understand the limitations of single source ascertainment, we described decedents who would not have been identified by a strategy using either VA decedent files alone or death certificates alone. We compared cases that would have been missed using either single source with the rest of the cohort based on their demographic and clinical attributes and the settings in which they received care. Methods In order to evaluate the implications for improving veterans' end-of-life care, we constructed a population-based decedent cohort [ 16 ]. For such purposes, it is particularly important to understand whether death was recorded elsewhere for veterans who were under VA care since the VA system may be responsible for much of their end-of-life care even if they do not die while receiving health care in a VA facility. Data Sources The VA Beneficiary Identification and Records Locator System (BIRLS) contains records of all beneficiaries including veterans whose survivors applied for burial benefits. It includes records of discharged military veterans post-1973 and recipients of Medals of Honor and VA education benefits. After submission to the Veterans Benefits Administration (VBA), deaths are recorded in the BIRLS Death File. A submission to the VBA is typically triggered by a family claim for death benefits (e.g. burial assistance, pension) [ 17 - 19 ]. The VA maintains a National Patient Care Database (NPCD) that contains a record of Social Security Number (SSN) linked VA and contracted health services provided to all veterans [ 17 - 19 ]. Death certificates are required for burial in California and are available for public use [ 20 ]. We first identified 345,380 decedent veterans who died during FY2000 (30 September 1999 – 1 October 2000) from the BIRLS Death File. We used SSNs to link cases to VA NPCD outpatient, inpatient, or long term care records restricted to recipients of any VA services in California within 12 months of death. We extracted records including any inpatient or long term care admission, or outpatient encounters. Veterans who entered the cohort on the basis of using outpatient services were required to have at least one clinical encounter (e.g., other than laboratory, radiology, or administrative). In addition, we used California death certificates as second source to identify decedent veterans by linking SSNs from death certificates directly to VA utilization files. California death certificates contained 462,561 records for calendar years 1999 and 2000, and we primarily matched decedents identified through death certificates to BIRLS by SSN. We manually inspected matches on SSN only and we also examined matches on criteria other than SSN (e.g. last name, first name, date of birth, date of death). Additional cases we accepted after manual inspection involved transpositions of one and rarely more than one SSN digit but agreement in other fields. Thus, the cohort included recipients of VA clinical services verified as deceased based on either BIRLS or death certificates, and all cases were linked to VA utilization files by SSN. In the final decedent cohort, we excluded cases of non-veterans receiving care at VA facilities by examining indicators of veteran status associated with visits. The VA assigns specific codes to non-veterans rendered care for various reasons (e.g., emergency or charitable care). We also considered the possibility of erroneous decedent status by looking for evidence of healthcare utilization during the 12 months after death. We excluded cases with evidence of utilization more than one month after the date of death. Variables and Analysis We used VA encounters and ICD-9-CM codes to demographically (e.g., age, gender, marital status, state of residence, and race / ethnicity) and clinically characterize decedents [ 21 - 26 ]. We identified veterans with any visit or admission for congestive heart failure (CHF), ICD-9-CM 398.91, 402.x1, 404.x1, 404.x3 428.x excluding procedures, chronic obstructive lung disease (COPD), ICD-9-CM 491–492.x, 494.x, 496, end-stage liver disease (ESLD), ICD-9-CM 571.2–571.9,572.2–572.8, dementia, ICD-9-CM 046.1, 290.0–290.43, 331.0–331.7, 333.4, 438.0, and malignant neoplasia, ICD-9-CM 140.0–208.9 [ 25 ]. To identify end-stage renal disease (ESRD), we used procedure and clinical stop codes that identify the type of care received (e.g., dialysis) [ 26 ]. We developed a complexity index of co-morbidity based on a simple count of advanced illnesses. To understand the limitations of single source mortality ascertainment, we described decedents who would not have been identified by a strategy using either death certificates alone or VA decedent files alone. We compared these cases with the rest of the cohort based on their demographic and clinical attributes and the settings in which they received care. Based on distributions, we used Wilcoxon tests for continuous and chi-square tests for categorical variables. Results From 345,380 deaths during the period 30 September 1999 to 1 October 2000 identified in BIRLS, we distinguished 6,071 decedents who were users of VA inpatient, outpatient, or long term care services in California. California death certificates included 227,308 deaths during the same period, including 3,580 additional users of VA inpatient, outpatient, or long term care services in California. Using SSN and other identifiers to match decedent cases to VA utilization data, we excluded non-veterans (n = 365), users of only non-clinical care such as laboratory tests (n = 251), those possibly alive based on subsequent VA encounter data (n = 229), and 3 cases for other reasons. Of the final cohort of 8,813 veteran decedents, 5,698 (65%) cases were identified in both source files, while 689 (8%) were only identified in VA decedent files, and 2,426 (28%) additional cases were only identified through death certificates (Figure 1 ). Figure 1 Cohort Development We examined potential biases associated with veteran decedents missed by either single source of mortality ascertainment (e.g., VA BIRLS or California death certificates). Ninety-nine percent of decedents missed by using VA data alone were California residents (vs. 92% of the remainder cohort, p < 0.001); whereas, 62% of those missed by using death certificates alone were out-of-state residents (vs. 1% of the remainder cohort, p < 0.001). Relatively fewer veterans of white or black ethnicity and relatively more veterans of missing ethnicity were represented among decedents missed by either single source strategy. The proportion of married or previously married veterans was higher and single or missing marital status lower among those decedents missed using only BIRLS, and relative proportions were reversed for a strategy using only death certificates. Decedents missed by either single source approach were less likely to have been diagnosed with an advanced chronic illness than the identified cohort. Veteran decedents missed by using only BIRLS were less likely to be diagnosed with any condition except HIV and dementia, and those missed by using death certificates alone were less likely to be diagnosed with any condition except HIV. With a BIRLS only approach, 37% of missing cases vs. 35% of the remainder cohort (p < 0.001) had no diagnosed chronic illness (death certificate only approach; 69% vs. 32%, p < 0.001). Veteran decedents missed by either single source approach were equally or more likely to have been users of the outpatient setting, but missed cases were less likely to have been users of inpatient healthcare settings (Table 1 ). Table 1 Potential Bias Associated with Alternative Strategies For Veterans' Mortality Ascertainment * BIRLS Only Strategy Death Certificate Only Strategy Cases identified by BIRLS Additional cases identified by death certificates P-value Cases identified by death certificates Additional cases identified by BIRLS P-value Number of cases 6,387 2,426 8,124 689 Age (years) 70.86 71.15 0.8253 70.95 70.79 0.6891 Gender Male 98 97 0.2733 98 98 0.8662 Race / Ethnicity White 57 54 58 31 <0.001 Black 12 8 11 6 Hispanic 5 5 5 1 Other 2 2 2 1 Missing 24 31 <0.001 23 61 Marital Status Married 46 49 47 45 0.0028 Single 16 13 15 16 Divorced 23 24 23 23 Widowed 11 12 11 11 Missing 4 2 0.0002 3 6 State of Residence California 92 99 99 38 Non-California 8 1 <0.001 1 62 <0.001 Diagnosis Cancer 35 32 0.0426 35 17 <0.001 CHF 22 19 0.0175 22 7 <0.001 COPD 28 24 0.0002 28 11 <0.001 ESLD 6 4 0.0327 6 3 0.0010 ESRD 3 1 <0.0001 3 0 0.001 Dementia 11 11 0.9656 11 3 <0.001 HIV 1 1 0.1586 1 0 0.0621 Complexity Index 0 35 37 32 69 1 35 39 38 22 2 22 18 22 6 3 7 5 7 2 4 1 1 <0.0001 1 0 <0.001 Site of Utilization Any inpatient Any long term 45 29 <0.0001 42 21 <0.001 care 20 12 <0.0001 19 7 <0.001 Any outpatient 95 96 0.0225 95 96 0.1218 *Findings are expressed as proportions unless otherwise identified. P-values reflect Wilcoxon two-sided probabilities for continuous variables and chi-square for categorical variables. Categorical tests reflect tests for differences including missing. Discussion Veterans' mortality ascertainment was significantly improved by using both VA and death certificates as source files. Our findings indicate that either single source approach for mortality ascertainment may misrepresent veteran mortality based on comparisons of race / ethnicity, marital status, severity of illness, and settings of care. Diagnoses associated with serious medical co-morbidity and the likelihood of receiving any inpatient services (e.g. hospital or long term care) were both significantly lower among veterans missed by either single source approach. Our findings are consistent with Washington State where the deaths of 25% of 533 veterans who only used outpatient services were only identified with death certificates, and 5% were only identified in BIRLS. [ 9 ] Using BIRLS only for mortality determination, it is unclear why generally healthier, primarily outpatient users are less likely to be noted. Death notification is typically triggered by benefit claims (e.g., burial assistance, pension and related benefits). Affluent veterans whose families might be less likely to file benefit claims were drawn to the VA recently [ 27 ]. However, poverty or low social support might also make it harder to file claims. On the other hand, a death certificate only approach to ascertainment misses relatively fewer non-resident veterans. Such veterans may be homeless or mobile, retired veterans, and they may seek care transiently in California, or their deaths may be recorded elsewhere. One limitation of our study is that we did not identify cases that were only decedents by virtue of VA utilization files alone rather than BIRLS, although Dominitz, et. al., identified only 2.7% of deaths this way [ 11 ]. We did not compare VA files or death certificates to the National Death Index (NDI), as have previous studies that have used the NDI as a gold standard. The NDI is a central data repository of state vital statistics that is often used as a gold standard in US mortality studies [ 28 ]. We report findings for only one state, but given similar findings in Washington State, it would be helpful to determine if this is a national issue or there are particular state issues related to BIRLS death file agreement, or concerns related to veteran morality ascertainment with California death certificates. Conclusion Researchers, managers, and policy makers should understand the limitations of sources of mortality ascertainment. The relationship of missing data to bias is related somewhat to how "missingness" is distributed by the outcome of interest. Our findings suggest these concerns may be relatively more important for studies involving veterans and racial-ethnic disparities, co-morbidity, certain disease comparisons, or settings of care. Additional study is needed to compare BIRLS, death certificates, and the NDI for mortality ascertainment in veterans. If our findings are confirmed, the VA may need to consider improving its system for mortality ascertainment through routine linkages to national mortality data. Studies of end-of-life care using decedent cohorts need to pay particular attention to the incompleteness of VA data as the sole source of mortality information. List of Abbreviations Used VA, Veterans Administration; FY, fiscal year; BIRLS, Beneficiary Identification and Records Locator System; NPCD, National Patient Care Database; SSN, Social Security Number; CHF, congestive heart failure; COPD, chronic obstructive lung disease; ESLD, end-stage liver disease; ESRD, end-stage renal disease. Competing interests The author(s) declare that they have no competing interests. Authors' contributions KL originated and oversaw all aspects of the conception, design, analysis, and publication of the study. SA, LR, and EY contributed to conception, design, and analysis. MW contributed to analysis and is responsible for programming. All authors reviewed and approved of the manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC554976.xml
539230	Unique Double-Barreled Enzyme Makes Methionine the Hard Way	null	If a cell is a complex symphony of chemical reactions, its enzymes are the instruments through which this elemental music is played. Each reaction is catalyzed by a specific enzyme, whose uniquely shaped active site not only binds reactants, but, by forming weak and temporary bonds, coaxes them into new orientations with new partners, thus creating the products. Determining exactly how any individual enzyme accomplishes this task—which amino acids make up the active site, which bonds form where when enzyme meets substrate, which electrons switch partners as new bonds form—is the work of the structural biochemist. In this issue, Martha Ludwig and Robert Pejchal elucidate the structure of cobalamin-independent methionine synthase (MetE) from the bacterium Thermotoga maritima , and describe how it catalyzes the formation of the amino acid methionine. Methionine synthases actually come in two forms, which use somewhat different mechanisms to accomplish the same task: transfer of a methyl group (CH3) from methyltetrahydrofolate to the terminal sulfur of homocysteine. The cobalamin-dependent form, MetH, relies on the cofactor cobalamin (vitamin B12), which pulls the methyl away at one active site, and then donates it at a second active site. Here, a central zinc atom binds and activates homocysteine, enabling it to attack the incoming methyl group that is attached to cobalamin. MetE, on the other hand, has no cofactor and only one active site, which sits at the junction of two eight-stranded barrels. The structure and sequence of these barrels indicate they arose through duplication of a primordial zinc-bearing, homocysteine-binding protein. This unique duplex now bears only one zinc atom, deep within the cleft separating the two barrels. As in MetH, the role of the zinc is to bind homocysteine, but in MetE, this event also induces a conformation change around the zinc. The zinc and its coordinating partners form an umbrella; entering from the handle end, the homocysteine sulfur pulls the zinc toward it and turns the umbrella inside out. Methyltetrahydrofolate initially binds along the edge of the cleft, with the methyl group on the folate oriented far from the sulfur on the homocysteine, as can be seen in the research article's Video S1 (DOI: 10.1371/journal.pbio.0030031.sv001 ). There must be subsequent conformational changes within the active site that serve to bring the two substrates together and promote transfer of the methyl group. Exactly how methyltetrahydrofolate reorients within the cleft to complete the reaction is not yet clear. The reaction catalyzed by MetE proceeds slowly, at only 1%–2% of the speed of that catalyzed by MetH. One reason for this rather sluggish activity is that homocysteine, even when activated by binding to zinc, is much poorer than cobalamin at displacing the methyl group of methyltetrahydrofolate. While MetE's unique active-site structure was made possible by gene duplication, the two barrels are no longer identical. Through evolution, the second, N-terminal, barrel has lost the ability to bind zinc or homocysteine, and indeed appears to contribute little to the active function of the enzyme. Nonetheless, this barrel may be necessary to temporarily isolate the substrates from solvent and to form the hydrophobic environment in which the reaction is more favorable. Further research may indicate more about the function of this unequal partner, and provide more detail on the exact atomic movements within the cleft at the moment of reaction.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC539230.xml
535813	Widespread presence of "bacterial-like" PPP phosphatases in eukaryotes	Background In eukaryotes, PPP ( p rotein p hosphatase P ) family is one of the two known protein phosphatase families specific for Ser and Thr. The role of PPP phosphatases in multiple signaling pathways in eukaryotic cell has been extensively studied. Unlike eukaryotic PPP phosphatases, bacterial members of the family have broad substrate specificity or may even be Tyr-specific. Moreover, one group of bacterial PPPs are diadenosine tetraphosphatases, indicating that bacterial PPP phosphatases may not necessarily function as protein phosphatases. Results We describe the presence in eukaryotes of three groups of expressed genes encoding "non-conventional" phosphatases of the PPP family. These enzymes are more closely related to bacterial PPP phosphatases than to the known eukaryotic members of the family. One group, found exclusively in land plants, is most closely related to PPP phosphatases from some α- Proteobacteria , including Rhizobiales , Rhodobacterales and Rhodospirillaceae . This group is therefore termed Rhi zobiales / Rh odobacterales / Rh odospirillaceae - l ike ph osphatases, or Rhilphs. Phosphatases of the other group are found in Viridiplantae , Rhodophyta , Trypanosomatidae , Plasmodium and some fungi. They are structurally related to phosphatases from psychrophilic bacteria Shewanella and Colwellia , and are termed She wanella - l ike ph osphatases, or Shelphs. Phosphatases of the third group are distantly related to ApaH, bacterial diadenosine tetraphosphatases, and are termed A paH- l ike ph osphatases, or Alphs. Patchy distribution of Alphs in animals, plants, fungi, diatoms and kinetoplasts suggests that these phosphatases were present in the common ancestor of eukaryotes but were independently lost in many lineages. Rhilphs, Shelphs and Alphs form PPP clades, as divergent from "conventional" eukaryotic PPP phosphatases as they are from each other and from major bacterial clades. In addition, comparison of primary structures revealed a previously unrecognised (I/L/V)D(S/T)G motif, conserved in all bacterial and "bacterial-like" eukaryotic PPPs, but not in "conventional" eukaryotic and archaeal PPPs. Conclusions Our findings demonstrate that many eukaryotes possess diverse "bacterial-like" PPP phosphatases, the enzymatic characteristics, physiological roles and precise evolutionary history of which have yet to be determined.	Background Reversible phosphorylation of proteins is a ubiquitous mechanism, indispensable for regulation of virtually any cellular function. Therefore, protein kinases and phosphatases are of paramount importance for normal functioning of all metabolic and signalling pathways. In eukaryotes, PPP family is one of the two known protein phosphatase families specific for Ser and Thr [ 1 - 4 ]. Unlike eukaryotic (and archaeal [ 5 ]) PPP phosphatases, bacterial members of the family have broad substrate specificity [ 6 ] or may even be Tyr-specific [ 7 - 9 ]. Moreover, one group of bacterial PPPs are diadenosine tetraphosphatases [ 10 , 11 ]. Unlike eukaryotes, in prokaryotes PPP phosphatases appear to be facultative, since entirely sequenced genomes of some bacteria and archaea do not encode them [ 5 , 12 ]. Nevertheless, when present, they appear to play essential roles [ 13 - 15 ]. Three motifs (GDXHG, GDXXDRG and GNH(E/D)), highly conserved in the N-terminal subdomains of the catalytic domains of all PPP phosphatases [ 10 , 11 ], contain most residues which coordinate metal ions in the active centre [ 16 ] and are considered as the signature of the PPP family. In a previous work [ 17 ] we identified an unusual cDNA fragment from a moss Physcomitrella patens , showing no similarities to the known PPP phosphatases beyond the presence of the GDXHG and GDXXDRG motifs. Detection of homologous cDNA sequences from Arabidopsis and rice suggested the presence of an unknown PPP group in plants, distinct from "conventional" eukaryotic PPP phosphatases [ 17 ]. We have now taken advantage of a much greater representation (as compared to 1999) of sequence databases for various species to further explore this initial observation. We present the evidence for the existence in eukaryotes of three "non-conventional" branches of the PPP family. We also identify a previously unrecognised conserved motif in the PPP catalytic domain, which can be used as a signature of "bacterial"-type PPP phosphatases. Results " Rh izobiales / Rh odobacterales / Rh odospirillaceae – like" PPP phosphatases in plants Two Arabidopsis sequences, At3g09960 and At3g09970, were retrieved using the P. patens fragment [ 17 ] as a query in TBlastN searches. They share 85% identity with each other at the protein level (see Figure 1 ). Both genes are transcribed, since full-length cDNAs have been detected in a large-scale transcription study [ 18 ]. They are arranged on chromosome 3 in tandem, suggesting their origin by a recent duplication. A number of ESTs from other plant species (but none from non-plant eukaryotes) were also detected by TBlastN searches, which in most cases provide no evidence for the existence of more than one isoform. Figure 1 Comparison of the primary structures of plant Rhilphs, related α-proteobacterial phosphatases and human PP1α as a prototype of "conventional" eukaryotic PPP phosphatases . Amino acid residues conserved in at least all but one Rhilphs and α-proteobacterial phosphatases are shown in white and shaded in black. Residues conserved in at least two thirds of the sequences are shown in white and shaded in dark grey. Residues conserved in at least half of the sequences are shown in black and shaded in light grey. Following substitutions were considered as conserved residues: Ile/Leu, Phe/Tyr, Asp/Glu, Asn/Gln, Arg/Lys and Ser/Thr. Catalytic site residues that interact with metal ions are indicated by asterisks according to [20]. SAPNY motif in PP1, conserved in most eukaryotic PPP phosphatases, is double underlined. Solanum tuberosum sequence is translation of the EST entries BQ516856, BQ516857 and BI435517. Physcomitrella patens sequence is translation of the EST entry BQ039171. Other accession numbers are indicated in Table 1. Among prokaryotes, related sequences were detected in some α -Proteobacteria , the closest matches being with Rhizobiales , Rhodobacterales and Rhodospirillaceae . Therefore, we designate this group as Rhi zobiales / Rh odobacterales / Rh odospirillaceae -like phosphatases, or Rhilphs (Figure 1 ; see also Table 1 and Figure 4 ), and the two Arabidopsis genes products At3g09960 and At3g09970 as " R hizobiales - l ike" p hosphatases 1 (RLP1) and 2 (RLP2), respectively. Table 1 Species, accession numbers (UniProt, EMBL, NCBI or TIGR Gene Index) and common names (where available) of PPP phosphatase sequences shown in Figure 4. For A. thaliana sequences, gene numbers are also indicated. Sequence No. 67 is available from Chlamydomonas reinhardtii draft genome [65]. No. Accession Species name Rhilphs 1 Q9SR61; At3g09960 Arabidopsis thaliana 2 Q9SR62; At3g09970 Arabidopsis thaliana 3 BE034080 Mesembryanthemum crystallinum 4 BQ995369 Lactuca sativa 5 AW034786 Lycopersicon esculentum 6 BU875146 Populus balsamifera 7 BG457803 Medicago truncatula 8 AV425727 Lotus japonicus 9 BM731295 Glycine max 10 BF261816 Hordeum vulgare 11 BQ788728 Triticum aestivum 12 AL731641 Oryza sativa 13 CF670562 Pinus taeda 14 BQ039171 Physcomitrella patens Group I (α-Proteobacteria) 15 Q987U4 Mesorhizobium loti 16 Q8UA33 Agrobacterium tumefaciens 17 ZP_00054691 Magnetospirillum magnetotacticum 18 ZP_00015226 Rhodospirillum rubrum 19 ZP_00014771 Rhodospirillum rubrum 20 CAE28794 Rhodopseudomonas palustris 21 Q9ABQ8 Caulobacter crescentus 22 ZP_00051041 Magnetospirillum magnetotacticum 23 ZP_00093979 Novosphingobium aromaticivorans 24 Q92V37 Sinorhizobium meliloti Group VII (heterogeneous) 25 Q8YZT4 Anabaena sp. 26 Q9WZK1 Thermotoga maritima 27 O34205 Fervidobacterium islandicum 28 NZ_AABE01000101 Cytophaga hutchinsonii Group III (γ-Proteobacteria and bacteriophage λ) 29 P03772 Bacteriophage λ 30 P55798 E. coli PrpA 31 Q8VPE2 Salmonella typhimurium PrpA 32 Q8Z487 Salmonella enterica 33 P55799 E. coli PrpB Group IV (Firmicutes) 34 Q81YR3 Bacillus anthracis 35 Q9FB69 Lactococcus lactis 36 Q97FF3 Clostridium acetobutylicum Alphs 37 BM291808 Amblyomma variegatum 38 TC9835 Ciona intestinalis 39 BU652795 Chlamydomonas reinhardtii 40 AC091781 Trypanosoma brucei 41 AC084046 Trypanosoma brucei 42 AL499620 Leishmania major 43 BQ143558 Metarhizium anisopliae 44 AC127427 Magnaporthe grisea 45 AA966318 Aspergillus nidulans 46 P40152 Saccharomyces cerevisiae ApaH 47 Q8Y1K9 Ralstonia solanacearum ApaH 48 Q9JVF4 Neisseria meningitidis ApaH 49 P05637 Escherichia coli ApaH Group VI (heterogeneous) 50 O31614 Bacillus subtilis 51 O69213 Anabaena sp. PrpA 52 Q93JF4 Streptomyces coelicolor 53 Q9RS78 Deinococcus radiodurans Group II (Cyanobacteria) 54 O54390 Microcystis aeruginosa PP1-Cyano 1 55 P74150 Synechocystis sp. 56 Q8YP31 Anabaena sp. 57 ZP_00072257 Trichodesmium erythraeum 58 Q8DGA2 Thermosynechococcus elongatus Shelphs 59 AC119500* Leishmania major 60 Q8EBN0 Shewanella oneidensis 61 Q9S427 Shewanella sp. 62 TIGR_167879 Contig1731 Colwellia psychrerythraea 63 CF394707 Pinus taeda 64 TC31593 Solanum tuberosum 65 Q944L7; At1g18480 Arabidopsis thaliana 66 BF645180 Medicago truncatula 67 Scaffold_45 Chlamydomonas reinhardtii 68 Q9LMJ5; At1g07010 Arabidopsis thaliana 69 AW266595 Mesembryanthemum crystallinum 70 BG644111 Lycopersicon esculentum 71 BG450922 Medicago truncatula 72 BI787505 Glycine max 73 Q8L676 Oryza sativa 74 TC21958 Hordeum vulgare 75 AC007863 Trypanosoma brucei 76 AL499621 Leishmania major 77 TIGR_246197 Contig433 Myxococcus xanthus 78 EAK84303 Ustilago maydis 79 O74480 Schizosaccharomyces pombe 80 EAK87480 Cryptosporidium parvum 81 Q7RIH8 Plasmodium yoelii 82 Q8IKE5 Plasmodium falciparum 83 Q8I5Y5 Plasmodium falciparum 84 Q7RR22 Plasmodium yoelii Group V (heterogeneous) 85 O87639 Streptomyces coelicolor 86 Q9RVT7 Deinococcus radiodurans Archaea 87 O28453 Archaeoglobus fulgidus PPA 88 Q8ZW26 Pyrobaculum aerophilum 89 O34200 Methanosarcina thermophila PP1-arch2 90 Y12396 Pyrodictium abyssi "Conventional" eukaryotic PPP 91 Q9U493 Plasmodium falciparum PPJ 92 Q8I728 Trypanosoma cruzi PPEF 93 BH900132 Ostreococcus tauri 94 O14829 Homo sapiens PPEF1 (PP7) 95 Q8IDE7 Plasmodium falciparum PP5 96 P53041 Homo sapiens PP5 97 P53043 Saccharomyces cerevisiae PPT 98 P32838 Saccharomyces cerevisiae PPG 99 P05323 Homo sapiens PP2A 100 O00743 Homo sapiens PP6 101 Q08209 Homo sapiens Calcineurin (PP2B) 102 P08129 Homo sapiens PP1 103 P32945 Saccharomyces cerevisiae PPQ 104 O49346 Arabidopsis thaliana PP7 * This phosphatase is unlikely to be catalytically active due to replacements of essential residues in the active centre. Figure 2 Comparison of the primary structures of " Shewanella -like" phosphatases (Shelphs) and human PP1α as a prototype of eukaryotic PPP phosphatases . Designations for conserved amino acid residues are as in Figure 1. For Oryza sativa , dashed underlined C-terminal sequence has been corrected by comparison with ESTs. Accession numbers: Plasmodium falciparum 1, Q8I5Y5; 2, Q8IKE5; Trypanosoma brucei 1, AC007863; 2, AC084046.12. Chlamydomonas reinhardtii sequence is translation of the EST entries BG855683 and BI995255. Other accession numbers are indicated in Table 1. Figure 3 Characteristic modifications (shaded in black) in the conserved PPP signature motif GDXXDRG in bacterial diadenosine tetraphosphatases and eukaryotic Alphs . Eukaryotic species are shown in bold. Plus signs indicate that gene expression is confirmed by the presence of ESTs. Figure 4 Neighbor-Net analysis of the conserved N-terminal subdomains (starting 5 amino acid residues before conserved GDXHG and ending 25 residues after GNH(E/D) of 104 bacterial, archaeal and eukaryotic PPP phosphatases . Bootstrap values exceeding 50% (out of 1000 resamplings) were obtained in a separate neihbour-joining analysis and are shown in brackets. Species and accession numbers are listed in Table 1. Note that groups designated as I, IV and VII did not receive significant bootstrap support; corresponding sequences are grouped together for convenience of their representation in Table 1. This image (and bootstrap values for the alternative splits) can be viewed at a higher resolution as the Additional File 1. Figure 5 Distinct conserved motifs in the C-termini of bacterial and "bacterial-like" PPP phosphatases from eukaryotes as opposed to archaeal and eukaryotic PPP phosphatases . A His residue directly binding a metal ion in the catalytic centre (marked with asterisk), and the elements of secondary structure are shown for bacteriophage λ phosphatase and for human PP1 according to ref. [20] and [19], respectively. The (I/L/V)D(S/T)G motif, highly conserved in bacterial and "bacterial-like" phosphatases, is highlighted. An expanded version of this alignment can be viewed as the Additional File 2. Structural features of Rhilphs All residues that are expected to bind metal ions in the catalytic centre are conserved in Rhilphs (Figure 1 ). Rhilphs do not have N- or C-terminal extensions beyond their catalytic domains, which in many PPP phosphatases have regulatory function and / or interact with regulatory proteins / subunits. Instead, they have characteristic inserts between the conserved motifs GNH(E/D) and HAG (corresponding to HGG in "conventional" eukaryotic PPPs, see Figure 1 ). Notably, much shorter inserts are found at a similar position in α-proteobacterial phosphatases (group I in Figure 1 ). Inserts in both plant Rhilphs and α-proteobacterial phosphatases contain a conserved motif LXXAXPXXH (Figure 1 ). Similarly to bacteriophage λphosphatase (λPP [ 19 ]), Rhilphs lack a region corresponding to β8, β9 and α9 of eukaryotic PPPs [ 20 - 22 ]. Like bacterial PPP phosphatases, Rhilphs do not have a SAPNY motif, conserved in the β12-β13 loop of eukaryotic PPPs. Analysis of Rhilph sequences did not reveal targeting or signal peptides. While this work was in progress, a phosphatase encoded by an Arabidopsis gene At1g07010 was reported in an independent study [ 4 ]. " Shewanella -like" PPP phosphatases in plants, red algae, fungi and unicellular parasites We undertook further TBlastN searches using full-length Arabidopsis RLP2 as a query to see whether Arabidopsis genome encodes additional "bacterial-like" PPPs. These searches identified two more genes for putative PPP phosphatases, only distantly related to Rhilphs and to any other members of the family (Figure 2 ). One of these genes is At1g07010 1 . At least three different predicted products of this gene could be found in protein databases. On the basis of comparison with EST sequences, we consider as the correct structure that of Q8RY10 with Asp and Gly at positions 109 and 208, respectively (see Figure 2 ). The other detected gene, At1g18480 , is also represented in the databases by three distinct deduced proteins. Comparison with A. thaliana ESTs confirms Q944L7 as the correct structure. Genomic and EST database searches provided ample evidence for the presence of related phosphatases in a number of green plants, including multiple angiosperm species, pine and a unicellular green alga Chlamydomonas reinhardtii (Figure 2 ; see also Table 1 and Figure 4 ). Related sequences were also identified in some fungi (several basidiomycetes and an ascomycete Schizosaccharomyces pombe , but not other ascomycetes), in Apicomplexa, Trypanosomatidae , and in a red alga Porphyra yezoensis (for available sequence from the latter species, see Figure 5 ). The most closely related prokaryotic phosphatases were detected in Myxococcus xanthus (δ- Proteobacteria ) and psychrophilic bacteria Alteromonadales (γ- Proteobacteria ): uncharacterised phosphatases from Shewanella oneidensis and Colwellia psychrerythraea and a Tyr-specific phosphatase PPI from Shewanella sp . [ 8 ]. Therefore, we designate this phosphatase group as " She wanella - l ike" ph osphatases, or Shelphs, and the products of the two prototype Arabidopsis genes At1g07010 and At1g18480 as " S hewanella - l ike" p hosphatases 1 (SLP1) and 2 (SLP2), respectively. Structural features of Shelphs Like in Rhilphs, all residues that are expected to bind metal ions in the catalytic centre are conserved in Shelphs (see Figure 1 ). Another feature common with Rhilphs is the presence of inserts (as compared to "conventional" eukaryotic PPPs) between the GNHE and H(A/G)G motifs (Figure 1 ), which are especially long in plant Shelphs. However, these inserts share no sequence similarity between Rhilphs and Shelphs and probably appeared in the two phosphatase groups independently. Like in Rhilphs, a region corresponding to β8, β9 and α9 of eukaryotic PPPs is absent in Shelphs, and the primary structure of the region corresponding to the β12-β13 loop is similar to that of typical bacterial PPPs. A.thaliana SLP1 and corresponding Shelph isoform from rice have chloroplast targeting sequence, which could not be detected in A.thaliana SLP2 and corresponding isoform from Medicago truncatula . Eukaryotic PPPs distantly related to bacterial diadenosine tetraphosphatases Identification of Rhilphs and Shelphs prompted us to perform extensive searches of eukaryotic sequence databases. These searches revealed the existence of other "bacterial-like" PPP phosphatases throughout eukaryotes. Sequences only distantly related to Rhilphs, Shelphs or any other PPP phosphatases were detected in several fungi (including a putative S. cerevisiae phosphatase reported previously [ 23 ]), in Trypanosomatidae , a tick Amblyomma , an ascidian Ciona , Chlamydomonas , pine and diatoms Fragilariopsis cylindrus and Phaeodactylum tricornutum . Blast searches using these sequences revealed that all of them share higher similarity to bacterial diadenosine tetraphosphatases (ApaH) than to other PPP groups. Therefore, we tentatively designate them as A paH- l ike p hosphatases, or Alphs (Figures 3 and 4 ; Table 1 ; partial sequences available for pine and diatoms are shown in Figure 5 ). Alphs share a distinctive common structural feature. In the GDXXDRG motif, absolutely conserved in other PPPs, the second Asp (which stabilises the protonation of a His that directly participates in catalysis [ 20 ]) is replaced by a neutral amino acid, and the Arg residue (which coordinates phosphate [ 24 ]) is replaced, with one exception, by Lys (Figure 3 ). The former of these replacements is also found in ApaH, while the latter is unique to Alphs. While higher overall sequence similarity and a common alteration in the GDXXDRG motif are compatible with closer relatedness of Alphs to bacterial diadenosine tetraphosphatases, phylogenetic analysis using full length sequences failed to produce a robust tree due to high sequence diversity (not shown). Relationship of novel eukaryotic PPP groups to known PPP phosphatases In order to better understand the relationship of "bacterial-like" PPP phosphatases in eukaryotes to each other and to bacterial PPPs, we attempted to extend our previous phylogenetic analysis of eukaryotic PPP phosphatases [ 25 ] by including PPP sequences from a number of bacteria and archaea. Primary structures of bacterial PPP phosphatases are extremely diverse and, outside the relatively conserved N-terminal subdomain of about 100 amino acids containing the GDXHG, GDXXDRG and GNH(E/D) motifs, they share only a few conserved residues. Moreover, many of the sequences have long insertions at different positions. This leads to the failure to produce informative alignments of full-length catalytic domains. Therefore, we aligned more conserved N-terminal subdomains only, an approach applied previously by Kennelly [ 6 ] to a much smaller set of PPP sequences available at that time. Phylogenetic reconstruction was attempted with either neighbor-joining (as implemented in PHYLIP [ 26 ] or SplitsTree [ 27 ]) or maximum likelihood analysis using quartet puzzling (TreePuzzle [ 28 ]); in the latter case a smaller dataset consisting of with consisting of 32 representative sequences was analyzed due to the inability of the algorithm to handle large datasets. Due to the relatively short length of the sequences and their high diversity, some of the major clades did not receive significant bootstrap support and were different depending on the method used, although most major clades, including Rhilphs and Shelphs, were recovered by both methods. Alphs tended to be grouped together by neighbor-joining but were split into smaller clades when maximum likelihood analysis was used. However, we still tentatively consider Alphs as a single group due to the characteristic replacements in their catalytic centre. To circumvent the ambiguity of the results, we used Neighbor-Net [ 29 ], a neighbor-joining based method that constructs phylogenetic networks rather than trees and thus represents conflicting signals and visualises feasible trees in a single plot (Figure 4 ; for a high-resolution image, see Additional file 1 ). The Neighbor-Net analysis accurately identified the major clades such as eukaryotic and archaeal phosphatases, as well as their closer relationship to each other than to bacterial PPPs [ 6 ]. Separation of "conventional" eukaryotic PPPs into two branches, suggested previously from the analysis of the full-length catalytic domains [ 25 ], was also recovered. As it was suggested by initial sequence similarity searches, Rhilphs, Shelphs and Alphs represent distinct major clades of the PPP family, as divergent from "conventional" eukaryotic and archaeal PPP phosphatases as they are from major bacterial clades (Figure 4 ; Additional file 1 ). Common structural elements in all "bacterial-like" PPP phosphatases from eukaryotes and bacterial phosphatases C-terminal regions of the catalytic domain of all "conventional" eukaryotic PPP phosphatases share a highly conserved (with minor variations) SAPNY motif, located in the β12-β13 loop. This loop and the Tyr residue of the SAPNY motif in particular are implicated in interaction with regulators and inhibitors [ 21 , 30 - 32 ]. β strands (β9 and β10) corresponding to β12 and β13 are conserved in λPP [ 19 ]. However, the sequence on the C-terminal side of β9 is dissimilar to SAPNY in λPP and in bacterial PPPs. A conservative replacement of the first Ser of SAPNY by Thr is found in many bacterial sequences (this Thr is however only moderately conserved and is replaced by Glu or Gln in all Rhilphs and by Val or Phe in most Shelphs). The two adjacent positions are occupied by highly conserved Asp and Gly residues, respectively, thus defining a previously unrecognised motif (I/L/V)D(S/T)G. This motif is present in all examined bacterial PPPs, as well as in all "bacterial-like" phosphatases from eukaryotes described above (Figure 5 ; see also Additional file 2 for a more complete alignment). In addition, we note the presence of another characteristic feature of bacterial and "bacterial-like" PPPs: the His residue (H248 in PP1; H186 in λPP) coordinating one of the metal ions in the catalytic centre is preceded by an absolutely conserved Gly; this residue is conserved in some archaeal but not in "conventional" eukaryotic PPPs (Figure 5 ; Additional file 2 ). Discussion In this report, we have documented the presence in different eukaryotic lineages of the genes that encode PPP phosphatases resembling those of bacterial origin, rather than "conventional" eukaryotic members of the family. Catalytic domains of these "bacterial-like" phosphatases are characterised by relatively conserved structure of the N-terminal subdomains, but very diverse organisation of the C-terminal subdomains, where conserved motifs and residues forming the active centre are separated by sequences of various length that share little or no similarity between different clades (Figure 6 ). In most cases, corresponding EST sequences could be detected, which confirms that these genes are expressed. Figure 6 Schematic diagram depicting organisation of the catalytic domains of the phosphatase groups discussed in this study . N-terminal subdomains (used in the alignment for the Neighbor-Net analysis, Figure 4) and C-terminal subdomains are shown in red and yellow, respectively. Positions of the conserved motifs in PPλ and the residues forming the active centre (underlined; shown according to ref. [19]) are shown. Positions of the LXXAXPXXH motif in plant Rhilphs and related phosphatases from Rhizobiales are indicated (green boxes). For more detailed information on the position of inserts in Rhilphs and Shelphs relative to the elements of the secondary structure, see Figures 1 and 2, respectively. The most conspicuous presence of "bacterial-like" PPPs has been detected in plants. Plants possess phosphatases from all three novel groups described in this work. Rhilphs are most closely related to PPP phosphatases from a number of α-proteobacteria, including purple photosynthetic bacteria and Rhizobiales . The absence of related sequences in eukaryotes other than land plants suggests that Rhilphs may have been acquired after plants started colonising land. Although bacterial lineage that could be the source for plant Rhilphs could not be unambiguously identified by phylogenetic reconstruction, Rhizobiales (or purple photosynthetic bacteria from which Rhizobiales are thought to have originated [ 33 , 34 ]) appear to be likely candidates. Indeed, Rhizobiales have transmissible chromosomal elements and some, like Agrobacterium , are able to integrate their plasmid genes into plant genome [ 35 ] or even transform animal cells [ 36 ], a situation that would be ideally suited for a horizontal gene transfer to occur. Interestingly, the presence of genes of rhizobial origin has been detected in plant parasitic nematodes [ 37 ]. Possible origin of plant Rhilphs from α-proteobacterial phosphatases is also supported by the presence in the enzymes of both groups of characteristic inserts in similar positions, which share some sequence similarity (see Results). Phosphatases of another group, designated as Shelphs, are found in green plants, in a red alga, in Apicomplexa , Trypanosomatidae , as well as in some fungi. The similarity between proteins from Apicomplexa and Trypanosomatidae and those from plants is well documented. Trypanosomatidae are related to photosynthetic euglenoids and are thought to have lost plastids secondarily [ 38 ]. Apicomplexan parasites have a relict plastid, originated from the engulfment of a red alga [ 39 ]. Thus, the presence of phosphatases shared by plants, red algae, Apicomplexa and Trypanosomatidae is not surprising and probably reflects the presence of Shelphs in a common ancestor of photosynthetic eukaryotes. The presence of chloroplast targeting sequence in SLP1 suggests a possible origin of Shelphs from a bacterial precursor of the chloroplast (however it should be noted that Shelphs are absent from cyanobacteria); alternatively, this sequence may have appeared secondarily. Protein Ser/Thr phosphorylation / dephosphorylation is essential for regulation of photosynthesis, and unidentified okadaic acid-insensitive protein phosphatases in chloroplasts have been reported [ 40 , 41 ]. SLP1 appears to be a good candidate for such a phosphatase. The origin of fungal Shelphs is unclear. Curiously, they are found in basidiomycetes and in an ascomycete S. pombe , but not in a number of other ascomycetes, whose genomes have been completed. Current data do not permit to discriminate between (i) the presence of Shelphs in a common ancestor of eukaryotes and their loss in such lineages as animals and many fungi, and (ii) independent acquisition of Shelphs from bacteria by an ancestor of photosynthetic eukaryotes and by fungi. Further sequencing of eukaryotic genomes may shed light on the evolutionary history of this PPP group. The third group of "bacterial-like" phosphatases detected in eukaryotes, designated here as Alphs, appears to be distantly related to bacterial diadenosine tetraphosphatases ApaH. Patchy distribution in several eukaryotic kingdoms suggests that Alphs were probably present in the common ancestor of eukaryotes, but were independently lost in many lineages, including insects, vertebrates and flowering plants. A characteristic modification of the conserved GDXXDRG motif shared only with ApaH further supports a suggestion that Alphs may represent a divergent branch of diadenosine tetraphosphatases, rather than protein phosphatases. However, relatedness of eukaryotic Alphs to bacterial diadenosine tetraphosphatases remains hypothetical, since Alph sequences are too divergent from ApaH, as well as from each other, to permit a reliable phylogenetic reconstruction. Diadenosine oligophosphates are considered as emerging signalling molecules in both intra- and intercellular signalling in eukaryotes [ 42 , 43 ]. In particular, human diadenosine oligophosphate hydrolase FHIT has been identified as a tumor suppressor [ 44 ]. It seems plausible that appearance of eukaryotic diadenosine oligophosphate hydrolases (structurally unrelated to the PPP phosphatases) may have made bacterial-type diadenosine tetraphosphatases redundant, leading to their loss in many eukaryotic lineages. It would be interesting to test experimentally whether Alphs are indeed diadenosine oligophosphatases. More generally, an important implication of our findings is that many eukaryotes possess PPP phosphatases with yet undetermined substrate specificity. Eukaryotic PPP phosphatases are generally considered as Ser/Thr specific in vivo , although they may be able to dephosphorylate phosphoTyr-containing substrates in vitro ( e.g . [ 45 , 46 ]). This is probably true for archaeal PPPs as well [ 5 ]. However, Ser/Thr specificity is not a feature of bacterial PPP phosphatases [ 7 - 9 , 13 , 47 - 49 ]. Thus, it would not be possible to predict substrate specificity of uncharacterised "bacterial-like" PPP phosphatases without experimental evidence. In particular, since Shewanella PPI is Tyr-specific [ 8 ], it would be interesting to determine substrate specificity of eukaryotic Shelphs. It is also worth noting that interest in tyrosine phosphorylation in plants has recently been stimulated by identification of plant Tyr phosphatase genes and by the finding that Tyr phosphorylation is involved in the regulation of stomatal movement (reviewed by Luan [ 50 ]). Three motifs, GDXHG, GDXXDRG and GNH(E/D) form the diagnostic signature of all PPP phosphatases [ 10 , 11 ]. We detected a (I/L/V)D(S/T)G motif, which appears to be a characteristic signature of "bacterial"-type PPPs. The existence of such a motif is striking per se , taking into account extreme structural diversity of bacterial PPP phosphatases. It indicates that (I/L/V)D(S/T)G was probably present as the fourth "universal" signature motif in the common ancestor of PPP phosphatases, and was lost in the common lineage of archaeal and "conventional" eukaryotic PPPs. An alternative possibility could be that the (I/L/V)D(S/T)G motif was acquired by a bacterium and propagated by lateral gene transfer, replacing the ancestral SAPNY-related motif. However this scenario seems less likely, since the (I/L/V)D(S/T)G motif is present, with minor variations, in virtually all bacterial phosphatases, despite their great diversity, and is replaced by SAPNY-related sequences only in archaeal and "conventional" eukaryotic PPPs. The Asp residue in the 2 nd position of (I/L/V)D(S/T)G is highly conserved and can only be replaced by Glu, indicating that the negative charge is essential. The presence of a highly conserved Gly in the 4 th position indicates that flexibility of the polypeptide chain is likely to be important. The crystal structure of bacteriophage λ phosphatase (PPλ; [ 19 ]) shows that the Asp residue of (I/L/V)D(S/T)G (Asp202) is just downstream of the β9 strand, which corresponds to the β12 strand in mammalian PP1. In PPλ, Asp202 is hydrogen bonded to a water molecule coordinated to one of the metal ions in the catalytic centre, which probably accounts for its conservation. In eukaryotic or archaeal PPPs, corresponding position is occupied by neutral residues (see Figure 5 ). It would be tempting to speculate that this difference in the region just downstream of the β9 (β12) may be responsible for a feature that is common to all "bacterial"-type but not to eukaryotic / archaeal PPPs. One such feature is the Ser/Thr specificity of the latter group. The Tyr residue of the SAPNY motif has been suggested to provide a bulky phenol ring in the β12-β13 loop, sufficient to sterically block access of phosphoTyr-containing substrates to the active site [ 32 ]. However this is unlikely to be the sole determinant of Ser/Thr specificity, since residues containing bulky aromatic rings (Tyr, Phe or Trp) are found in the same or adjacent positions in many bacterial phosphatases (Figure 5 ). Since the (I/L/V)D(S/T)G motif, absent in eukaryotic and archaeal PPPs, is involved in organisation of the catalytic centre [ 19 ], it is possible that this difference in the catalytic centre organisation may be one of the determinants of broad substrate specificity vs . Ser/Thr specificity. Conclusions So far, eukaryotic PPP phosphatases were considered as a well-defined monophyletic group of enzymes, specifically dephosphorylating phosphoSer and phosphoThr, while a much more structurally and enzymatically diverse PPP phosphatases were known to be present in prokaryotes. Our findings demonstrate that, in addition to "conventional" eukaryotic PPP Ser/Thr-specific protein phosphatases, many eukaryotes possess very diverse "bacterial-like" PPP phosphatases. Enzymatic characteristics, physiological roles and evolutionary history of these phosphatases have yet to be revealed. Methods Detection of PPP phosphatase-coding sequences Sequence similarity searches were conducted using BlastP or TBlastN [ 51 ] at NCBI [ 52 ] in the following databases: "non-redundant" (NR), "expressed sequence tags" (EST), "genomic sequence survey" (GSS) and "high-throughput genomic sequences" (HTGS). Additional Blast searches of the following databases were performed: finished and unfinished genomes of eukaryotes at the NCBI [ 53 ]; fungal genomes at the Broad Institute [ 54 ]; plant genomes at The Arabidopsis Information Resource (TAIR) [ 55 ]; Gene Index databases of tentative consensus sequences (EST assemblies) at The Institute for Genomic Research (TIGR) [ 56 ]; Chlamydomonas reinhardtii draft genome [ 57 ]. In all cases, reciprocal searches were used, i.e. hits retrieved by Blast searches were in their turn used as queries in the following Blast searches. Accuracy of gene prediction was examined by comparison of the retrieved sequences with translations of corresponding EST entries. In the absence of available ESTs, closely related sequences from other species were used. Taxonomy of the species from which the phosphatase sequences is given according to the NCBI taxonomy web site [ 58 ]. Phylogenetic analysis Multiple alignments were generated using CLUSTAL W [ 59 ] at Kyoto University Bioinformatics Centre [ 60 ] and edited manually. During manual editing, particular attention was paid to correct alignment of the PPP family signature motifs and other conserved residues known to constitute the catalytic site of PPP phosphatases. Phylogenetic tree construction by the neighbor-joining method [ 61 ] and bootstrap analysis were performed using the PHYLIP package, version 3.573 [ 26 ]. Maximum likelihood analysis was performed using TreePuzzle [ 28 ]. Possible alternative neighbor-joining based phylogenies were visualised using Neighbor-Net [ 29 ] as implemented in SplitsTree, version 4.β10 [ 62 ] Analysis of the primary structure The presence of signal peptides and targeting sequences was analyzed using TargetP [ 63 ] at the the Centre for Biological Sequence Analysis, Technical University of Denmark [ 64 ]. Abbreviations EST, expressed sequence tag, Alph, ApaH – like phosphatase; PPP, protein phosphatases of the P family; Rhilph (RLP), Rhizobiales / Rhodobacterales / Rhodospirillaceae – like phosphatase; Shelph (SLP), Shewanella – like phosphatase. Authors' contributions Both authors contributed equally to this work. Supplementary Material Additional File 1 Neighbor-Net analysis of the conserved N-terminal subdomains (starting 5 amino acid residues before conserved GDXHG and ending 25 residues after GNH(E/D) of 104 bacterial, archaeal and eukaryotic PPP phosphatases . This version of Figure 4 is the original SplitsTree file that can be viewed using SplitsTree, freely available for download (see Methods). Bootstrap values (out of 100 resamplings) are shown and can be highlighted by selecting corresponding alternative splits. The file also contains the alignment used for the analysis (Input). Click here for file Additional File 2 Distinct conserved motifs in the C-termini of bacterial and "bacterial-like" PPP phosphatases from eukaryotes as opposed to archaeal and eukaryotic PPP phosphatases . This is an expanded version of Figure 5 . Click here for file	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC535813.xml
535344	Bi-directional modulation of AMPA receptor unitary conductance by synaptic activity	Background Knowledge of how synapses alter their efficiency of communication is central to the understanding of learning and memory. The most extensively studied forms of synaptic plasticity are long-term potentiation (LTP) and its counterpart long-term depression (LTD) of AMPA receptor-mediated synaptic transmission. In the CA1 region of the hippocampus, it has been shown that LTP often involves a rapid increase in the unitary conductance of AMPA receptor channels. However, LTP can also occur in the absence of any alteration in AMPA receptor unitary conductance. In the present study we have used whole-cell dendritic recording, failures analysis and non-stationary fluctuation analysis to investigate the mechanism of depotentiation of LTP. Results We find that when LTP involves an increase in unitary conductance, subsequent depotentiation invariably involves the return of unitary conductance to pre-LTP values. In contrast, when LTP does not involve a change in unitary conductance then depotentiation also occurs in the absence of any change in unitary conductance, indicating a reduction in the number of activated receptors as the most likely mechanism. Conclusions These data show that unitary conductance can be bi-directionally modified by synaptic activity. Furthermore, there are at least two distinct mechanisms to restore synaptic strength from a potentiated state, which depend upon the mechanism of the previous potentiation.	Background Fast excitatory synaptic transmission in the central nervous system, which is mediated predominantly by the AMPA subtype of glutamate receptors, can undergo long-term bi-directional modifications in strength [ 1 - 3 ]. These persistent changes have been proposed to be key synaptic processes involved in learning and memory. The best characterised form of bi-directional synaptic plasticity is LTP / LTD of glutamatergic transmission in the CA1 region of the hippocampus. Although there has been intensive investigation of the induction and expression of LTP and LTD (see [ 4 ]), the precise molecular mechanisms by which these alterations in synaptic strength occur remain unclear. Possible mechanisms could be presynaptic such as changes in the release process (probability of neurotransmitter release or the amount of L-glutamate released from vesicles (e.g., [ 5 - 9 ]), and / or postsynaptic (e.g., [ 10 - 12 ]), such as a change in the number of AMPA receptors (AMPARs) available to bind transmitter, or in the properties of existing receptors (P open , activation or de-activation kinetics, desensitisation or single-channel conductance, γ). Recent studies have provided information on the possible mechanisms underlying postsynaptic alterations in synaptic strength. There is evidence that AMPA receptors are inserted into the postsynaptic membrane during LTP [ 13 - 15 ] and removed from the synapse upon induction of LTD [ 16 , 17 ]. However, it has also been shown that LTP can involve a rapid increase in γ of existing AMPA receptors [ 18 ]. This could be caused by a Ca 2+ /calmodulin-kinase II (CaM-KII)-mediated phosphorylation of GluR1 which occurs during LTP [ 19 ], and which causes an increase in γ of GluR1 homomers in transfected cells [ 20 ]. Therefore, another potential mechanism for LTD could be a decrease in γ caused by a dephosphorylation of AMPARs. We have recently used peak-scaled non-stationary fluctuation analysis (non-SFA; [ 21 ]) of synaptic currents recorded from CA1 pyramidal cell dendrites [ 18 ], to investigate the molecular basis of de novo LTD (LTD at naïve pathways; [ 22 , 23 ]). In these studies, LTD was never associated with a change in γ [ 17 ]. Indeed, evidence was presented that the underlying mechanism involved a reduction in the number of surface expressed AMPA receptors (LTD N ). In the present study we have investigated a second form of LTD known as depotentiation (DP), which is a reversal of pre-established LTP [ 24 - 27 ]. LTP can be associated with either an increase in γ (LTPγ) or no change in γ (LTP N ) [ 18 ]. We were, therefore, interested to determine whether DP of LTPγ involved a decrease in γ and hence whether γ is a bi-directionally modifiable parameter. We find a reciprocal relationship between LTP and DP, such that DP of LTPγ invariably involves a restoration of the pre-LTP γ (DPγ) whereas DP of LTP N never involves a change in γ (DP N ). These data show, firstly, that there are two distinct molecular mechanisms for the reduction of synaptic strength that are dependent on the nature of the preceding LTP and, secondly, that γ can indeed be bi-directionally modified in response to synaptic activity. Results Using whole-cell recordings from the proximal apical dendrites of hippocampal CA1 pyramidal cells, minimal stimulation of nearby afferents evoked EPSCs that could be reliably resolved from failures (trials in which stimulation produced no synaptic response; Figure 1 , see also Figure 3 ). These high resolution recordings enabled both a failures analysis to be performed [ 28 ] and, using non-SFA, an estimate of γ of synaptically-activated AMPARs to be obtained [ 18 ]. LTP To investigate the mechanism of DP, LTP was first induced in 18 cells by pairing afferent stimulation (baseline frequency) with a holding potential of 0 mV. This resulted in stable LTP (EPSC amplitude = 186 ± 16 % of baseline, n = 18). In agreement with a previous study [ 18 ], cells fell into two groups with respect to changes (> 20%) in γ of AMPA receptor channels during LTP. In the majority of cases (11/18 cells), LTP was associated with an increase in γ (LTPγ; 246 ± 26% of baseline; range 136 – 363%). In the other 7 cells there was no increase in γ during LTP (100 ± 4 % of baseline; range 85 – 118 %), indicating that there was an increase in the functional number of channels activated (LTP N ). As noted previously [ 18 ], there were no differences between the two groups of neurons with respect to a variety of baseline parameters. Depotentiation of LTPγ Figure 1 shows two examples from the group of cells that exhibited LTPγ, as indicated by the change in the current-variance plot obtained from non-SFA (Figure 1B ). As previously reported [ 18 ], the increase in γ was not associated with any change in EPSC kinetics (Figure 1C ) indicating that AMPA receptor channel kinetics were not affected [ 29 ]. Failures analysis (Figure 1C ) of this group of cells (Figure 2 ) revealed that LTP was associated with changes in success rate (1 – failure rate) in some cells (Figure 2B ), and potency (mean EPSC amplitude excluding failures) in all cells (Figure 2C ), as previously reported under these recording conditions [ 18 , 28 ]. For this group of cells, DP, induced by pairing stimulation (baseline frequency) with a holding potential of -40 mV, always resulted in a reversal of the γ increase, as indicated by the current-variance plot (Figure 1B ; Figure 2D ). Similar to the preceding LTPγ, this form of DP (DPγ) was also associated with no change in EPSC kinetics (τ rise : baseline = 1.6 ± 0.2 ms, LTPγ = 1.7 ± 0.3 ms, DPγ = 1.7 ± 0.2 ms, n = 11; τ decay : baseline = 8.3 ± 0.6 ms, LTPγ = 8.2 ± 0.6 ms, DPγ = 8.9 ± 0.7, n = 11; Figure 1C ). Failures analysis showed that DPγ was associated with a decrease in success rate in most cells (Figure 2B ) and a full reversal of the potency increase (Figure 2C ). These data show that the primary mechanism for DP is the reversal of any increase in γ caused by LTP. Indeed, the changes in γ were sufficient to account for the potency changes during both LTPγ and DPγ. (In most cells the alterations in γ actually exceeded the potency changes. This is most likely due to an underestimate of the potency change due to dendritic filtering, which affects measurements at the peak of EPSCs greater than during the tail, from where the non-SFA estimates are obtained; see [ 18 ]). Depotentiation of LTP N Figure 3 shows an example from the group of cells that exhibited no change in γ during LTP (LTP N ), as indicated by the current-variance plot (Figure 3B ). The change in EPSC amplitude for LTP N neurons (Figure 4A ) was similar to that for LTPγ neurons (Figure 2A ). Failures analysis (Figure 3D ) of this group of cells (Figure 4 ) revealed that LTP was associated with changes in success rate in some cells (Figure 4B ), and potency in most cells (Figure 4C ; P < 0.01), as previously reported under these recording conditions [ 18 , 28 ]. In contrast to DPγ, DP in these cells was never associated with a change in γ (DP N ; Figure 3B , Figure 4D ). There was also no change in EPSC kinetics with LTP or DP (τ rise : baseline = 1.9 ± 0.4 ms, LTP N = 2.0 ± 0.3 ms, DP N = 1.8 ± 0.3 ms, n = 7; τ decay : baseline = 9.7 ± 0.7 ms, LTP N = 9.7 ± 0.6 ms, DP N = 10.1 ± 0.9, n = 7; Figure 3C ). Failures analysis of LTP N and DP N (Figure 3D , Figure 4 ) showed similar changes to the LTPγ group of cells in EPSC amplitude (Figure 4A ), success rate (Figure 4B ) and potency (Figure 4C ). These data suggest that there is a second mechanism for DP, not involving a decrease in γ, which co-exists at CA1 synapses. Therefore, there are two mechanisms for the expression of DP that depend upon the expression mechanism of the previous potentiation. Role of NMDA receptors in depotentiation It has been shown previously that DP at CA1 synapses may be blocked either by NMDA receptor antagonists [ 26 ] or by mGlu receptor antagonists [ 27 ] and that this may depend on previous history [ 30 ]. In the present study we wished to focus on NMDA receptor-dependent synaptic plasticity and therefore used pairing protocols designed to activate NMDA receptors sufficiently to, firstly, induce NMDA receptor-dependent LTP and, secondly, to induce NMDA receptor-dependent DP. To verify that we were indeed investigating NMDA receptor-dependent DP we performed a series of experiments using the NMDA receptor antagonist D-AP5 interleaved with control experiments. Following the induction of LTP, D-AP5 (50 μM) was bath applied for 15 minutes before delivering the DP induction stimulus. Whilst DP was induced in the control experiments (25 ± 11% of baseline, n = 5; p < 0.05) it was blocked by D-AP5 (89 ± 21%, n = 4; Figure 5 ). Relationship of changes in success rate, potency and γ to the magnitude of DP To gain further insights into the underlying mechanisms of DP we compared changes in EPSC amplitude, success rate, potency and γ for the individual experiments (Figure 6 ). A decrease in success rate indicates a reduction in probability of transmitter release (Pr) and/or a reduction in the number of functional synapses (n). A decrease in potency indicates a reduction in quantal amplitude (postsynaptic response to the release of a single quantum of transmitter, q) or, if multiple synapses are activated, a reduction in Pr or n. In both types of neuron (i.e., LTPγ and LTP N ), a small depression of less than 50 % (to 71 ± 7% of baseline; n = 4) was associated with no change in success rate (Figure 6A ; success rate ratio = 1.00 ± 0.01) but an equivalent decrease in potency (Figure 6B ; potency ratio = 0.70 ± 0.08), as is also observed for de novo LTD [ 17 ]). This indicates that the depression in these cells was associated primarily with a decrease in q. For larger depressions (to 26 ± 4 % of baseline; n = 14) there was also a marked decrease in success rate (success rate ratio = 0.52 ± 0.08; P < 0.01 vs success rate ratio for the group with DP < 50%) as well as a decrease in potency (potency ratio = 0.54 ± 0.04). This suggests that in these cells there was an additional decrease in Pr or n. Therefore, for DPγ, the change in γ did not fully account for the amplitude change in every cell (Figure 6C ) but did account for the potency change (Figure 6D ). Discussion In this study we have shown that there are two mechanisms for NMDA receptor-dependent DP, a reduction in γ (DPγ) and a decrease in the number of activated AMPA receptors (DP N ). As reported previously for different data sets from this age of rats [ 9 , 18 ] there are two forms of LTP; in approximately two-thirds of neurons LTP was associated with an increase in γ (LTPγ) whilst in the remainder there was no change in γ (LTP N ). In the present study we saw a similar proportion of LTP expressed by changes in γ versus N. Strikingly, we observed a precise relationship between the mechanism of DP and the form of preceding LTP; LTPγ was always reversed by DPγ, and LTP N was always reversed by DP N . In some experiments, induction of DP caused a decrease in EPSC amplitude below the initial baseline. This is most likely due to the simultaneous induction of DP and de novo LTD since in slices taken from juvenile animals, de novo LTD is readily induced by this [ 17 ] and other [ 22 , 23 ] induction protocols. This is in contrast to previous experiments using adult tissue in which DP induction depressed synaptic responses only as far as the initial baseline and where the same induction protocol was unable to induce de novo LTD [ 27 ]. Analysis of the mechanism of de novo LTD under the present experimental conditions demonstrated that it was associated with no change in γ [ 17 ]. Therefore the coexistence of DP and de novo LTD does not interfere with the analysis of DPγ. In addition, there is sometimes a small, gradual run-down of synaptic responses observed in minimal stimulation experiments using two-week-old animals [ 17 ] see also [ 31 ]. Whilst this effect tends to exaggerate changes in amplitude and success rate during DP in some neurons, it does not significantly interfere with estimates of potency or γ (see [ 17 ]). Mechanisms underlying DP N What might be the mechanism underlying DP that is not associated with a decrease in γ (i.e., DP N )? It is unlikely that this type of depression is due to a change in channel kinetics because there was no change in EPSC kinetics (see [ 18 , 29 ]). Therefore the mechanism is most likely a reduction in the number of activated AMPARs. This could be due to a presynaptic mechanism such as a reduction in release probability, the L-glutamate content of vesicles or the amount of L-glutamate discharged during fusion. Indeed there is evidence that some forms of LTD are expressed presynaptically [ 5 , 35 ]. Postsynaptic mechanisms for a reduction in the number of activated AMPA receptors include a reduction in their P open [ 36 ]or in the physical number of receptors present in the postsynaptic membrane [ 37 ]. The present observations for DP N are indistinguishable from those that we and others have reported recently for de novo LTD [ 17 , 38 ]. For example, we showed that similar effects were obtained using the postsynaptic injection of a peptide (pep2m) that disrupts the interaction between NSF and GluR2 [ 39 - 42 ]. The effects of pep2m and those of de novo LTD were mutually occlusive, indicating a convergence of mechanisms. These data argue strongly for a postsynaptic mechanism of expression. Furthermore, since pep2m causes the removal of AMPA receptors from the membrane surface, as determined immunocytochemically [ 38 , 42 ], it is most likely that de novo LTD is due to the physical elimination of synaptic AMPA receptors. Other evidence for a postsynaptic mechanism for de novo LTD includes a reduction in the postsynaptic sensitivity to glutamate [ 43 , 44 ], the dephosphorylation of serine 845 of the GluR1 subunit [ 45 , 46 ] and a rapid internalisation of AMPA receptors [ 47 ] associated with LTD. Therefore, by analogy, we feel that the postsynaptic removal of AMPA receptors is also the most likely explanation for DP N (Figure 7A ). Accordingly, a reduction in AMPA receptor number would account for the changes in potency without changes in success rate observed with modest DP N . The removal of an entire synaptic complement of AMPA receptors would explain the additional change in success rate seen with large depressions associated with DP N in some cells. Mechanisms underlying DP γ A number of possible underlying mechanisms could account for the change in γ during LTPγ and DPγ. Non-SFA cannot distinguish between 1) a change from a single low conductance state to a single high conductance state, 2) changes in open times within a burst and, 3) changes in the proportion of time spent in different conductance states. Since AMPA receptors are known to have multiple conductance states [ 32 , 33 ] we have postulated that the proportion of time spent in different conductance states is the modifiable parameter [ 18 ]. Such a change is detectable with the type of analysis that we have used, which provides a weighted mean of the various sub-conductance states [ 48 ]. Independent support for this hypothesis is provided by a study which shows that phosphorylation of GluR1 at serine 831 increases the proportion of time AMPA receptors spend in high conductance states, as determined by single channel recording [ 20 ]. Indeed, phosphorylation of this residue occurs during LTP [ 19 ]. Thus a possible mechanism of DPγ is the dephosphorylation of serine 831 [ 45 ], perhaps involving calcineurin [ 49 ], resulting in a lower proportion of time AMPA receptors spend in the higher conductance states (Figure 7B ). Other theoretical possibilities exist to explain DPγ. For example, the silencing of synapses close to the patch electrode leaving more distant synapses contributing a lower net γ due to electrotonic filtering, or a decrease in trial-to-trial asynchrony of transmitter release. We believe that these possibilities are unlikely because in every cell in which de novo LTD was induced there was never a change in γ [ 17 ]. If such possibilities were likely, statistically one would expect similar changes to occur for both de novo LTD and DP. Moreover, we have also investigated these possibilities using our standard compartmental model [ 29 ]. This shows that 1) the electrotonic filtering required to achieve an artefactual decrease in γ would have a pronounced effect on τ decay which was never observed experimentally, and 2) pronounced changes in asynchrony necessary to cause an artefactual change in γ cause large deviations from the parabolic relationship in the current-variance plot (unpublished observations) and alterations in τ rise [ 29 ], also never observed experimentally. Another possibility is that alterations in vesicle fusion pore dynamics leading to substantial changes in the peak and time-course of cleft glutamate are caused by synaptic plasticity [ 7 ]. Such changes could differentially affect estimates of γ [ 50 ], however they would be associated with substantial changes in τ rise and τ decay [ 7 ], which were never observed experimentally. Conclusions In summary, we have shown that there are two distinct molecular mechanisms for the reduction of synaptic strength. Although previous studies have provided evidence that LTD is associated with dephosphorylation of serine 845 on GluR1 [ 45 ] and internalisation of AMPA receptors [ 16 , 17 ] it is not known whether this represents two separate mechanisms or two components of the same process. For example, dephosphorylation of GluR1 could drive the internalisation of AMPA receptors. Here we show, for the first time, the co-existence of two distinct mechanisms for the expression of DP, using functional criteria under identical experimental conditions. The relationship between the two mechanisms is critically dependent upon the recent experience of the synapse, which may be governed by the phosphorylation state of the AMPA receptor complement [ 45 ]. Further work is now required to elucidate the relationship between these two fundamental mechanisms for modulating synaptic strength and the precise molecular mechanisms involved in each form of plasticity. Methods Electrophysiology Hippocampal slices (400 μm) were obtained from 12–15 day old rats and perfused with an extracellular solution containing (in mM): 124 NaCl, 3 KCl, 1.25 NaHPO 4 , 26 NaHCO 3 , 2 CaCl 2 , 1 MgSO 4 , 15 glucose, 2 ascorbic acid, 0.05 picrotoxin, saturated with 95% O 2 / 5% CO 2 , at room temperature (23–25°C). Individual dendrites were visualised using infrared illumination and DIC optics and approached under visual control. Whole-cell dendritic recordings of synaptic currents were obtained at a holding potential of -70 mV using patch electrodes (6–10 MΩ) filled with a solution containing (in mM): 135 CsMeSO 4 , 8 NaCl, 10 HEPES, 0.5 EGTA, 4 Mg-ATP, 0.3 Na-GTP, 5 QX-314, pH 7.25, 285 mOsm. Schaffer collateral-commissural fibers were stimulated at 0.5 Hz using a platinum monopolar or concentric bipolar electrode, which was positioned 20–40 μm from the dendrite parallel to the input pathway. The stimulus intensity was set to evoke some failures to enable a failures analysis to be performed and to ensure that the majority of EPSCs were, for any given trial, evoked by release from a single site. However, to elicit sufficient EPSCs to obtain a baseline estimate of γ before "washout" of LTP it was usually necessary to set the success rate fairly high (usually above 50%). As a result it is likely that multiple release sites contributed to the recordings. LTP was induced by pairing 40–60 stimuli (baseline frequency) with a holding potential of 0 mV. DP was induced following the induction of stable LTP by 100–200 stimuli (baseline frequency) at -40 mV holding potential. All recordings were made using an Axopatch 1B amplifier, signals were filtered at 5 kHz (8 pole Bessel filter), digitised at 10 kHz and stored on computer. EPSC amplitude and input resistance were analysed and displayed on-line using the 'LTP' program [ 51 ]). Series resistance was estimated by measuring the peak amplitude of the fast whole-cell capacitance current in response to a -1 mV step applied to the cell during each sweep. The amplitude was estimated by fitting the capacitance transient with a double exponential (from 0.5 ms after the peak) and determining the current at the beginning of the step. Series resistance was stable throughout recordings (series resistance values [MΩ]: baseline = 42 ± 3, LTP = 40 ± 3, DP = 43 ± 3, n = 17). Analysis Non-SFA was performed as described previously [ 18 ]. Briefly, synaptic currents were aligned by their point of maximal rise, and averaged. The average response waveform was scaled to the peak, subtracted from individual responses and the variance of the decays calculated. The variance was plotted vs . the mean current amplitude and the single channel current was estimated by fitting the data to: σ 2 = iI - I 2 /N + b l , where σ 2 is the variance, I is the mean current, N is the number of channels activated at the peak, i is the single channel current and b l is the background variance. The single channel conductance (γ) is then γ = i/V, where V is the driving force (holding potential – assumed reversal potential of 0 mV). Response amplitude was measured in two ways. For failures, amplitude was estimated by measuring the difference between the average current over two time windows of equal length, one immediately before the stimulus artefact and the other centred on the peak of the mean EPSC. For estimation of the amplitude of successes the peak amplitude over a set time window was determined. Failures were identified visually, and potency was calculated as the mean EPSC amplitude excluding failures. For analysis of EPSC kinetics, rise time was estimated by the time-constant of a single exponential fit of the rising phase of the mean EPSC waveform. For decay, the time-constant of the single exponential fit to the decay phase was used. For display of individual EPSC traces in the figures, the stimulus artefact was digitally subtracted using an average of identified failures. All histograms are represented as smoothed line plots (SigmaPlot ver 5.0). Data are expressed as % of baseline (i.e., 100 % = no change). Statistical significance was assessed using the Student's t -test (one or two-tailed, paired or unpaired as appropriate; P < 0.05 as significant). All values are expressed as mean ± s.e.m. Authors' contributions AL performed most of the electrophysiology. MJP and MAW also contributed some recordings. PM provided some information that guided the work. TB performed the modelling that provided the theoretical frame-work. JI helped with the design and analysis of experiments. GLC coordinated the work and contributed to its design. All authors read and approved the manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC535344.xml
544581	Metallopanstimulin as a marker for head and neck cancer	Background Metallopanstimulin (MPS-1) is a ribosomal protein that is found in elevated amounts in the sera of patients with head and neck squamous cell carcinoma (HNSCC). We used a test, denoted MPS-H, which detects MPS-1 and MPS-1-like proteins, to determine the relationship between MPS-H serum levels and clinical status of patients with, or at risk for, HNSCC. Patients and methods A total of 125 patients were prospectively enrolled from a university head and neck oncology clinic. Participants included only newly diagnosed HNSCC patients. Two control groups, including 25 non-smokers and 64 smokers, were studied for comparison. A total of 821 serum samples collected over a twenty-four month period were analyzed by the MPS-H radioimmunoassay. Results HNSCC, non-smokers, and smokers had average MPS-H values of 41.5 ng/mL, 10.2 ng/mL, and 12.8 ng/mL, respectively (p = 0.0001). Conclusion We conclude that MPS-1 and MPS-1-like proteins are elevated in patients with HNSCC, and that MPS-H appears to be a promising marker of presence of disease and response to treatment in HNSCC patients.	Background Effective therapy for head and neck squamous cell carcinoma (HNSCC), which constitutes approximately 95% of head and neck malignancies, is dependent upon early diagnosis and intervention. Despite the obvious advantage to earlier diagnosis of head and neck malignancies, no strategy has proven to effectively detect these tumors at early stages. Most head and neck neoplasms are detected when the patient has become symptomatic from the effects of the primary disease or when lymphatic metastases are palpable. These tumors are infrequently found incidentally on physical exam, and in these cases are often discovered at an earlier stage. Stage of disease at time of diagnosis is the primary metric used for determination of therapy and prognostication of life expectancy [ 1 ]. As tumor stage advances, the morbidity of surgical resection worsens due to an increased loss of tissue volume and involvement of vital structures. Organ-sparing approaches to head and neck malignancies have been developed in an attempt to treat advanced stage lesions while avoiding conventional surgical morbidities. They have, however, not produced universally superior results to surgery both in terms of local-regional control and function. Surveillance in the post-treatment head and neck cancer patients has traditionally centered on regular physical examination [ 2 , 3 ]. Office flexible fiberoptic exams of these patients have provided an excellent means of diagnosis for early mucosal recurrences, but are dependent upon the patient's compliance with regular follow-up and often cannot detect submucosal recurrence. Anatomic imaging is used as an adjunct to regular physical exam when recurrence is suspected, when findings are suggestive of cervical lymphatic involvement, or when a patient's symptoms are out of proportion or unexplained by physical exam findings. Imaging of anatomic structures is complicated by alterations in anatomy due to previous surgery or irradiation. Furthermore, despite many promising early reports, no tumor marker has yet been adopted for clinical use which shows high specificity or sensitivity for primary or recurrent HNSCC [ 4 - 8 ]. Metallopanstimulin-1 (MPS-1) was identified, cloned and characterized in the laboratory of Dr. Fernandez-Pol from a cDNA library constructed from a human mammary carcinoma cell line (MDA-468) that was stimulated by the growth factors TGF-β1 and EGF in the presence of cyclohexamide [ 9 ]. MPS-1, a multifunctional S27 ribosomal protein, is an 84 amino acid 9.5 kD ribosomal subunit, "zinc finger" protein that is present in all tissues and expressed in large quantities in a wide spectrum of proliferating tissues and oncogenic processes [ 10 - 16 ]. When MPS-1 is over-expressed, it is either secreted or passively released down a concentration gradient into the extra-cellular space. Conventionally, ribosomal proteins are thought to be confined in their function to intracellular protein synthesis. Many recent reports have drawn attention to "extraribosomal functions" of ribosomal proteins [ 17 - 20 ]. Moreover, these extraribosomal functions have been observed in relation to oncogenesis in various models [ 21 , 22 ]. The zinc finger motif of MPS-1 and other ribosomal proteins may allow binding to nucleic acids which may result in interference with transcription and translation [ 10 , 17 , 18 ]. Practical applications of this include: 1) DNA repair, 2) gene suppression, 3) cell-cycle control, or 4) control of oncogenesis. Another related ribosomal protein, S27a, is ubiquitinilated and over expressed in human colon cancer. Like MPS-1, it is involved in cell-cycle control and DNA replication [ 23 ]. The physiology of MPS-1 expression and our initial experience with this protein in HNSCC has led us to conclude that MPS-1 and MPS-1-like proteins may be useful markers in the effort to screen for and analyze the extent of HNSCC [ 24 ]. The purpose of this study was to use an empirical MPS-H test, which measures both MPS-1 and MPS-1-like proteins, to 1) compare average MPS-H levels between HNSCC patients and normal controls and 2) to illustrate how the MPS-H test may be useful for surveillance and evaluating response to treatment for HNSCC. Patients and methods Patients A total of 125 volunteers with newly diagnosed HNSCC were prospectively enrolled from a university head and neck oncology clinic. Serum collection consisted of a pre-treatment specimen followed by collections every six weeks during the first year and quarterly during the second year. At the time of specimen collection, presence of HNSCC was judged by all available data including: physical examination (that included an office endoscopy when indicated), biopsy, and radiology [computerized tomography (CT), magnetic resonance imaging (MRI), or positron emission tomography (PET)]. A clinical assessment was rendered based on all available data as "no evidence of disease (NED)", "alive with disease", or an indeterminate status. Serum MPS-H levels in these patients (N = 709) were compared to two control groups. The first control group was comprised of 25 normal, healthy, non-smoking volunteers. The second control group included 64 actively smoking volunteers who were screened for HNSCC and were found to be free of disease during the 1999 Yul Brynner Head and Neck Screening day in St. Louis, Missouri. A total of 821 serum samples collected over a twenty-four month period were analyzed by the MPS-H radioimmuno assay (follow-up days, mean: 217, median: 166.). All patients gave informed consent under an IRB-approved protocol. MPS-H Serum Assay Technical details for the preparation of reagents for MPS-H antigen determinations, RIA procedure, and patient sample preparation are published elsewhere [ 9 , 10 ]. Each serum sample was run in duplicate by a single technician who was blinded to specimen identity. The targets of this assay are the MPS-N-terminus of both MPS-1 and MPS-1-like proteins. These proteins are activated or released from the precursor or carrier proteins by heat-denaturing of the serum under controlled conditions. The resulting proteins are collectively designated as MPS-H. Immunoreactive substances detected by the MPS-H test do not reflect the true levels of authentic immunoreactive MPS-1/S27 ribosomal protein in the circulation under non-denatured conditions. Statistical Analysis Average MPS-H levels were compared between HNSCC patients and the two control groups and across American Joint Committee on Cancer (AJCC) stages and subsites using analysis of variance (ANOVA). Based on the results from the clinical assessment we computed the receiver operating characteristic (ROC) curve for MPS-H. The ROC curve plots the trade-off between sensitivity and specificity for a range of threshold values for defining a positive result. All ANOVA and contingency table analysis were performed using SAS (version 8.1, Cary, NC) statistical software. ROC curve analysis was performed using R software, an open source implementation of the S language (version 1.4.1, ). Statistical significance was defined as p < 0.05. Results Of the 125 HNSCC patients studied, 90 were male (Table 1 ). Table 2 presents the distribution of stage and site of the primary tumor. Most patients presented in stage III (24.0%) or IV (50.4%) with primary tumors of the oral cavity (26.4%) and larynx (37.6%). Table 1 Demographic and treatment characteristics of cases Gender Radiation/Chemo Surgery Total Female N 7 28 35 % 5.6% 22.4% Male N 19 71 90 % 15.2% 56.8% Total N 26 99 125 % 20.8% 79.2% 100.0% Table 2 Summary of primary tumor stage and location. Primary Stage Tumor Site I II III IV Total Percent Hypopharynx 0 1 0 1 2 1.6% Larynx 7 4 18 18 47 37.6% Nasopharynx 0 0 0 4 4 3.2% Neck 0 1 1 3 5 4.0% Oral Cavity 5 8 4 16 33 26.4% Oropharynx 0 0 2 18 20 16.0% Parotid 0 0 2 0 2 1.6% Skin 4 2 3 3 12 9.6% Total 16 16 30 63 125 Percent 12.8% 12.8% 24.0% 50.4% 100.0% MPS-H levels in this group were compared to a control group of healthy volunteers and to a control group who volunteered for screening for head and neck cancer (Figure 1 ). Mean MPS-H was 10.2 ng/mL for the healthy control group, and 12.8 ng/mL for the smoking control group. Mean MPS-H for the HNSCC group was 41.5 ng/mL, which was significantly higher than both control groups (p < 0.0001). Furthermore, Figure 2 illustrates that among HNSCC patients, those who were successfully treated and clinically free of disease had consistently lower MPS-H levels over time than patients living with active head and neck cancer. Figure 1 Mean serum MPS-H level for patients with SCC (n = 125, all stages and sites within the head and neck), healthy control group (n = 51) and actively smoking volunteers who were screened for HNSCC (N = 64). Mean of SCC group is 41.5 ng/mL, mean of healthy control group is 10.2 ng/mL, and mean of screening controls is 12.8 ng/mL (p < 0.0001). Error bars denote 1 standard deviation. Figure 2 Serial MPS-H levels in HNSCC patients treated and without clinical disease (No Disease) and patients with unresectable disease or receiving palliative therapy with persistent clinical disease (Alive with Disease). Error bars denote 1 standard deviation and vary widely in the AWD group due to its small size and patient attrition over time from death. We next computed the receiver operating characteristic (ROC) curve for the MPS-H levels. Figure 3 shows that the area under the ROC curve (0.73; 95% CI: 0.71–0.75; p = 0.001), is significantly different from 0.5, which suggests that there is moderate diagnostic accuracy associated with MPS-H. Analysis of MPS-H levels as a function of AJCC stage or head and neck sub site were performed and were not significant. Larger numbers of earlier stage (I and II) tumors and greater numbers among the various sub sites might result in significance. Figure 3 Receiver operating characteristic (ROC) curve for MPS-H test. Area under the curve is 0.73 (CI 95 : 0.69 – 0.76, p = 0.001) We have observed several instances where elevated MPS-H levels in patients presenting with head and neck neoplasms dropped to normal levels following successful therapy. We have also noted examples of persistent elevations or increases in MPS-H levels in patients with failure to respond to therapy or with recurrence of tumor respectively. Several cases of patients presented in Figures 4 , 5 , 6 illustrate these phenomenons. Patient 1 (Figure 4 ) was a female with a T 4 N 0 M 0 SCC of the floor of mouth with positive tumor margins on the cut edge of the mandible, having failed a recent limited surgery by another surgeon. Her presenting MPS-H value was borderline positive when she had little clinical disease [1 on the x axis] and rose immediately postoperatively. A transient rise after surgical ablation or induction chemotherapy is a documented phenomenon observed with numerous tumor markers (personal communication J.A. Fernandez-Pol). The elevation likely results from a large initial disruption of cells within the tumor resulting in a dumping of intracellular MPS-1 into the circulation. This usually returns to baseline in 4 to 6 weeks if the tumor is successfully treated. The patient was clinically NED in the immediate postoperative period and during radiation treatment. Approximately 6 months post operatively, the patient developed a neck mass and by the next visit skin metastasis were seen. On last follow-up, the patient had progression of skin and neck metastasis at which time she decided to pursue hospice care. She expired 6 weeks later. Figure 4 Patient without evidence of disease from surgery through radiation therapy (draws 1–5). The patient suffered a recurrence of clinical disease following radiation therapy (draws 6–8), which progressed until death. Figure 5 Patient followed for 24 months without evidence of recurrence on physical exam, endoscopy, or FDG-PET. Figure 6 Patient with unresectable tumor. Note MPS-H level response to chemotherapy. Patient subsequently expired after refusing further therapy. Patient 2 (Figure 5 ) is a male who was diagnosed with a T 4 N 2c M 0 SCC of the larynx, which required total laryngectomy, bilateral neck dissections, and postoperative radiation therapy. The patient joined this study 3 months following his surgery, while clinically NED, and has remained free of clinical recurrence for 24 months. The patient had three FDG-PET scans at six-month intervals following his surgery that were all negative for recurrent disease or metastasis. His persistently low MPS-H levels over time (range from 7.7–19.2 ng/mL), was suggestive of ongoing disease-free status. Patient 3 (Figure 6 ) was a female who presented with a T 4 N 0 M 0 SCC of the hypopharynx. She was severely malnourished with multiple medical problems and thought to be a poor surgical candidate. She decided to pursue induction chemotherapy (Carboplatin and Taxol) to be followed by radiation therapy. She underwent three rounds of chemotherapy at 21-day intervals and had a 50% decrease in her tumor size as judged by office endoscopy. She suffered severe GI problems during her therapy, opted not to continue on to radiation, and enrolled in hospice care. The patient did not present for additional follow-up after entering hospice and died three months later. This case illustrates the potential utility of MPS-H as a marker for tumor response to chemotherapy and/or irradiation. Discussion Histological examination of the tumor, surgical margins, and cervical nodes are the current means of determining extent of disease. When surgical extirpation is not undertaken, staging is performed based on a radiological assessment, biopsy, and physical examination. These methods are used either to determine adequacy of resection and the need for adjuvant therapy or to select an alternative primary (non-surgical) therapy, respectively. Limitations of the histologic method include microscopic disease that escapes diagnosis due to a small number of malignant cells, subtle histological changes not classified as cancer, previous treatment effect upon tissues, or pathologic sampling error. CT and physical examination both suffer from modest sensitivity and specificity in detecting many early head and neck neoplasms. As a result, local and regional treatment failures are not uncommon in both surgical and non-surgical treatments of head and neck cancer. This may be due to an underestimation of the tumor burden. This may also be explained by the current diagnostic emphasis upon analysis of structure (microscopic) or anatomic extent of disease, both of which are imperfect, rather than its biologic activity as might be measured by a tumor marker or a functional scanning technique such as FDG-PET. The MPS-H test has been used in conjunction with conventional tumor-specific markers to improve sensitivity and specificity of tumor serodiagnosis [ 12 ]. Of all malignancies in which MPS-H has been studied to date, epithelial malignancies possess no alternative tumor markers in clinical use that have been effective for diagnosis or surveillance [ 4 - 8 , 23 ]. This observation in conjunction with the current dependence on anatomic evaluation for diagnosis of epithelial malignancies has led us to preliminary investigations of the utility of MPS-H serologic diagnosis for the detection of head and neck epithelial neoplasms. Control groups of healthy volunteers and those with systemic, non-malignant diseases have been studied using the MPS-H test. Statistical analysis has revealed that those without malignancy have MPS-H serum levels less than 10 ng/mL, those with malignancy have levels greater than 20 ng/mL, and those with bony metastasis have levels in excess of 100 ng/mL [ 12 ]. Additionally, MPS-H levels have been documented to decline with successful treatment of malignant disease whereas non-responders to therapy persisted with high levels of MPS-H [ 12 ]. Serum samples from prostate carcinoma patients with high levels of MPS-H (>500 ng/mL) have demonstrated the authentic 9.4 kDa MPS-1 protein, at least one protein with sequence homology to the N-terminus of MPS-1, and a high molecular weight precipitation interfering protein. More accurate detection of recurrent SCC of the head and neck by the MPS-H test may provide a new way to improve survival. Physical exam, conventional imaging, and biopsy are the current gold standard to determine recurrence. Currently, FDG-PET is the most promising way to assess early tumor recurrence of the head and neck but is quite expensive [ 25 , 26 ]. Its sensitivity and specificity have been reported to be approximately 90% [ 25 , 26 ]. No standard serum tumor marker is routinely used for head and neck cancer surveillance, which limits alternatives to conventional exams or frequent FDG-PET imaging [ 4 - 8 , 25 ]. In the present study we compared FGD-PET to MPS-H levels in head and neck cancer patients. FDG-PET interpretation and MPS-H level determination were performed independently and blinded from the results of the other test. FDG-PET positive scans were not all confirmed by biopsy in our study. A statistically significant correlation was noted between FDG-PET positive cases and high MPS-H serum levels in head and neck cancer patients. MPS-H and FDG-PET agreed in 103 out of 183 cases. In 12 cases MPS-H was elevated but no cancer was found by FDG-PET, suggesting that the patients may have had an early recurrence detectable by MPS-H but not yet by FDG-PET. The 68 FDG-PET positive cases that show low MPS-H levels suggest that some tumors were unable to produce high levels of MPS-H, perhaps due to previous chemotherapy, or that some results represent false positive PET scans. Further study with larger patient groups is ongoing to assess the optimal cutoff levels of MPS-H and the correlation between FDG-PET and MPS-H. Shortcomings of this study include a limited time span, a large proportion of advanced stage cancers, limited controls (both size, smoking status, and age/gender match), and ability to define an absolute cut off value for normal vs. abnormal. We recognize that other factors such as age and other malignancies may effect MPS levels. Future studies will attempt to address and/or control for these issues. Considering the data presented in this paper, which agrees with previous results with other tumor types, and the compelling need to expedite the early diagnosis of primary and recurrent epithelial malignancies of the head and neck, we are further evaluating the MPS-H tests as a tool for diagnosis in a larger group of HNSCC patients. Additionally, we are working to improve the diagnostic technology used to detect MPS-H [ 27 - 29 ]. Since there is a reasonable correlation between detection of MPS-H in the sera and FDG-PET positivity for SSC, these results raises the potential of the MPS-H test for becoming a test for HNSCC, followed by selective confirmatory FDG-PET imaging. These preliminary results will be verified in a larger population. The role for using MPS-H as a general screening tool among at-risk populations without the diagnosis of HNSCC is also currently being evaluated. Competing interests The authors declare that they have no competing interests. Authors' contributions BCS: Principal Investigator, Principal Editor CSH: Data analysis and manuscript preparation CL: Medical student research assistant for project FD: Member of multidisciplinary head and neck oncology team, recruiter of subjects VJL: Member of multidisciplinary head and neck oncology team, supplier of PET data, data analysis. PDH: Laboratory technical support with MPS assay Funding sources This study was funded by the clinical salaries of BCS, FRD, and VJL. CML and CSH were students at the time of this study. MPS assays were performed at the Department of Veterans Affairs James A. Cochran Medical Center, Molecular Oncology Laboratory, St. Louis, MO. PDH was an employee of the US Government at the time of this research.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC544581.xml
524261	One Brain, One Vision	null	Not all devices that measure the same property do it in the same way—a clock might use a spring system or it might be digitally synchronized to a transmitted signal. Although both have the same goal of reporting accurate time, each is subject to different errors. Sometimes even the same device uses different systems to measure the same property. A relatively simple device like a camera will use one sensor system to capture light intensity for an image and a second sensor to capture light intensity for making automatic adjustments of aperture and flash. It does not seem outlandish, therefore, that the brain might also have developed multiple sensory systems to achieve different goals. Indeed, an influential hypothesis has argued that people use two separate visual processing systems in much the same way as a camera—one for creating our perception of the world and another for guiding our actions within it. One line of evidence supporting this dual hypothesis comes from an illusion known as Roelofs effect. Usually, people are pretty good at judging the location of even a small object. But if the small object is surrounded by a large frame and the frame itself is not centered in front of the person who is judging it, the viewer will perceive the object as shifted in a direction opposite that of the frame. This may not in itself be surprising, but the same person who perceives an offset of the object where none exists is nonetheless able to grasp it without difficulties. Figure 1 In this issue of PLoS Biology , Paul Dassonville and his colleagues reexamine the seeming dissociation of visual analysis for perception and action, and call it into question. Through a careful quantitative analysis of the conditions under which the Roelofs effect occurs, they find that it traces not to an illusory perception of the object location but to an illusory perception of self. The large frame, presented under experimental conditions in which subjects sit in darkness without access to a normal rich sensory environment, actually causes people to incorrectly perceive their own centers as rotated towards the frame and therefore to conclude that the small object is offset with respect to themselves. This may seem like a subtle distinction, and yet, since it is the observer's frame of reference that is altered, that same distorted frame of reference will be used to guide movement. Thus, the error in movement planning should cancel the error in perception, and people should have no trouble reaching for the object despite their misperception, which is indeed what is observed. Others have questioned the hypothesis that two separable neural systems process the visual world for perception and action, but this study removes one of the strongest pieces of evidence in its favor with a precise alternative explanation. No two brains may see the world identically, but the authors suggest that it may be time to concede that a single brain, at least, has the same world view.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC524261.xml
524513	An unusual case of chronic meningitis	Background Chronic meningitis is defined as symptoms and signs of meningeal inflammation and persisting cerebrospinal fluid abnormalities such as elevated protein level and pleocytosis for at least one month. Case presentation A 62-year-old woman, of unremarkable past medical history, was admitted to hospital for investigation of a four-week history of vomiting, malaise an associated hyponatraemia. She had a low-grade pyrexia with normal inflammatory markers. A CT brain was unremarkable and a contrast MRI brain revealed sub-acute infarction of the right frontal cortex but with no evidence of meningeal enhancement. Due to increasing confusion and patient clinical deterioration a lumbar puncture was performed at 17 days post admission. This revealed gram-negative coccobacilli in the CSF, which was identified as Neisseria meningitidis group B. The patient made a dramatic recovery with high-dose intravenous ceftriaxone antibiotic therapy for meningococcal meningitis. Conclusions 1) Chronic bacterial meningitis may present highly atypically, particularly in the older adult. 2) There may be an absent or reduced febrile response, without a rise in inflammatory markers, despite a very unwell patient. 3) Early lumbar puncture is to be encouraged as it is essential to confirm the diagnosis.4) Despite a delayed diagnosis appropriate antibiotic therapy can still lead to a good outcome.	Background Bacterial Meningitis usually presents as an acute illness, predominantly affecting children and young adults. It typically presents with the classical clinical triad of fever, neck stiffness and an altered mental state. However, it may rarely present as a chronic illness, without the classic clinical features noted in acute meningitis. Case presentation A 62-year-old retired woman was admitted to hospital via her GP for investigation of a four-week history of vomiting and malaise associated with hyponatraemia. She was initially diagnosed as suffering from viral gastroenteritis. However, the vomiting had persisted and had become associated with a mild frontal headache. She had an unremarkable past medical history and was not taking any regular medication. She had never smoked and there was no recent antecedent foreign travel. On examination she appeared clinically dehydrated but otherwise looked well, and was alert and orientated. She was apyrexial and had no rash, photophobia, neck stiffness or stigmata of endocarditis. She had a sinus tachycardia of 104/minute, with normal heart sounds, and a blood pressure of 130/76 mmHg. Chest, abdominal and neurological examinations were unremarkable. She had a plasma sodium of 127 mmol/L (135–145 mmol/L), potassium of 3.4 mmol/L (3.5–5.0 mmol/l), urea 4.8 mmol/L (3.0–6.5 mmol/L) and creatinine of 68 mmol/L (60–125 mmol/L). There was no biochemical evidence of the syndrome of inappropriate antidiuretic hormone (SIADH) production (serum osmolality 261 mmols/kg; urine osmolality 71 mmols/Kg; urine sodium < 10 mmol/L). Serum complement and plasma immunoglobulin levels were unremarkable with no evidence of immunosuppression. In addition, she had a normal full autoimmune profile and thyroid function. Random cortisol level was mildly elevated at 799 nmol/L (normal 140 – 700 nmol/L) consistent with a stress response. Her initial white cell count (WCC) was mildly elevated at 13.0 × 10 9 /L (normal 4–11 × 10 9 /L) with a neutrophilia of 10 × 10 9 /L (normal 2–7.5 × 10 9 /L). Her ECG and chest X-ray were normal. Her C-reactive protein (CRP) was slightly elevated at 10 mg/L (normal <5 mg/L) and erythrocyte sedimentation rate (ESR) was normal at 5 mm/hour. Her Chest X-ray and electrocardiogram were normal. Initial microbiological investigations (blood cultures, urine microscopy and culture) were normal. Initial management consisted of slow intravenous rehydration with normal saline and antiemetic therapy, which led to a mild symptomatic improvement. Upper gastrointestinal endoscopy revealed mild oesophagitis. During the ensuing two weeks her laboratory investigations remained stable (CRP normal; ESR normal; sodium 127–131 mmol/L; WCC 11–13 × 10 9 /L). However, on day 4 of admission she developed a low-grade pyrexia of 37.5°C, which persisted (<38°C). A CT scan of the head revealed periventricular patchy white matter changes but no features of raised intracranial pressure or space occupying lesion. Unfortunately the patient had become slowly more lethargic, withdrawn, and depressed. By day 17 of admission, although alert, she was uncooperative with intermittent confusion. Her symptoms of intermittent nausea and vomiting with occasional frontal headache continued. On day 18 she underwent a lumbar puncture (LP) as she still had a low-grade pyrexia (temperature 37.5°C) and neutrophilia of 9.3 × 10 9 /L). In addition, her nausea and vomiting had failed to fully settle with supportive treatment. The LP results were as follows: cerebrospinal fluid (CSF) appearance was pale yellow and clear; protein = 5.69 g/L (0.15–0.4 g/L); CSF glucose 1.7 mmol/L versus plasma glucose 5.7 mmol/L (ratio = 30%, normal > 50%); CSF WCC = 106/mL (normal <5 WCC/mL) – 99% lymphocytes. Gram's stain revealed gram-negative coccobacilli; acid-fast bacilli were not seen. She was commenced on intravenous ceftriaxone. Contrast MRI brain revealed sub-acute infarction of the right frontal cortex but with no evidence of meningeal enhancement. EEG demonstrated slow wave activity, which was consistent with a meningo-encephalitis. Within 48 hours of intravenous antibiotics she was more alert, orientated, and sitting out of bed. CSF culture grew gram-negative cocci, which was identified as Neisseria meningitidis group B, type NT, subtype NT P1.16/nt. She underwent contact tracing and completed a 10-day course of intravenous ceftriaxone. She continued to make a slow but progressive recovery. After a period of rehabilitation and intense physiotherapy she was discharged home 40 days after admission, with mild residual gait ataxia. Conclusions This case report presents two important clinical concepts: firstly, the presentation of chronic meningitis and secondly, the clinical presentation of bacterial meningitis in the older adult (defined as > 60 years old). The diagnosis was delayed due to the highly atypical clinical presentation [ 1 ]. Chronic meningitis is defined as symptoms and signs of meningeal inflammation and persisting cerebrospinal fluid (CSF) abnormalities such as elevated protein level and pleocytosis for at least one month [ 2 , 3 ]. It affects less than 10% of meningitis sufferers and is linked to a large variety of both infective and non-infective causes [ 4 ]. However, whilst there are numerous published individual case reports on chronic meningitis, there is a definite paucity of large case series in the literature. The most common cause of chronic meningitis is Mycobacterium tuberculosis, which accounts for 40–60% of cases [ 3 , 5 ]. Other relatively frequent causes include malignancy (8–13%) and crytococcal infection 7–11%) [ 3 , 5 ]. In up to 33% of cases no underlying cause is identified [ 3 , 5 ]. Chronic meningococcal meningitis is rare and is limited to a few isolated case reports in the literature [ 6 - 8 ]. There are several distinguishing features that may help to differentiate chronic meningitis from adult acute bacterial meningitis (table 1 ). The classic triad of clinical features of meningitis (fever, neck stiffness, altered mental state), whilst seen in up to 85% of patients presenting with acute bacterial meningitis is far less commonly seen in chronic meningitis [ 3 , 5 , 9 ]. Focal neurological signs with cranial nerve palsies and abnormal CT brain findings are also far more commonly seen in chronic meningitis [ 5 , 10 ]. Table 1 Features distinguishing chronic meningitis (bacterial and non bacterial) compared with acute bacterial meningitis Description Acute bacterial meningitis Chronic meningitis Aetiology Variable Neisseria meningitides 13–56% [10,17, 39] Streptococcus pneumoniae 24–37% [10,17] Variable TB- 40–60% [3,5] Malignancy 8–13% [3,5] Cryptococcus 7–11% [3,5] Unknown 30–33% [3,5] Clinical features - Classic triad of fever, headache and neck stiffness 85% [9] 10% [4] - Fever 78–91% [39] 44% [4] - Headache 32–68% [39] 79% [4] - Neck Stiffness 58–82% [39] 75% [5] - Altered Mental state 52–82% [39] 41% [4] - Focal neurology 23% [39] 32% [5] - Papilloedema <1–4% [9,10] 30% [5] - Cranial Nerve Palsies 4% [10] 24% [5] Mortality Variable – aetiology dependent 19.7–25% overall [10,17] 37–44% ≥ 60 years old [1,10] 10–25% < 60 years old [10,16–20] Variable – aetiology dependent 29%- overall [5] Elevated WCC, CRP and ESR Elevated Normal or only mildly elevated [5] Hyponatraemia <10% >90% [5] Cerebrospinal fluid analysis 10% – lymphocytic [9,17] 90% – neutrophilic [9,17] Gram stain positive 57–90% [9,10,17] >90% lymphocytic [5] <10% neutrophilic [5] Gram stain positive <10% [5] Abnormal CT 2.7 – 13% [10,40] 60% [5] WCC, white cell count; CRP, C-reactive protein; ESR, erythrocyte sedimentation rate. Hyponatraemia (as in our patient), whilst very uncommon in acute bacterial meningitis, is seen in the vast majority of cases of chronic meningitis [ 5 , 11 ]. Although there was a persistent mild neutrophilia, both the CRP and ESR were normal throughout the course of the disease, which, whilst being highly unusual for acute meningitis, has been reported, in chronic meningitis [ 12 , 13 ]. Acute bacterial meningitis is usually a rapidly progressive and highly lethal disease in older adults [ 1 ]. Rapid diagnosis is vital as the prognosis worsens with treatment delay leading to a high rate of sustained neurological deficit in this age group [ 14 , 15 ]. Despite the widespread use of antibiotics the overall case mortality rate remains unchanged and is far higher (37–44%) in the older adult compared with that seen in younger adults (10–25%) with significant long-term morbidity (up to 70% of infected patients) in survivors [ 16 - 20 ]. Given the success of childhood immunization, and an increasingly aging population, the proportion of older adults presenting with bacterial meningitis is increasing [ 16 ]. There are several additional factors, which make the older adult more prone to bacterial meningitis. Older adults often have underlying acute and chronic diseases (e.g. diabetes, renal or hepatic failure) with immunosenescence (age related decline in immune function) [ 1 , 21 ]. This can lead to symptoms, which can be confused with those of meningitis and at the same time increase the propensity to infection [ 1 , 21 ]. The role of immunosenescence in predisposing patients to bacterial meningitis is not clearly defined, but appears to relate to defects in innate, specific cellular and humoral immunity leading to an attenuated immune response [ 1 , 22 - 24 ]. Persons who lack or are deficient of antibody-dependent, complement-mediated lysis (bacteriocidal activity) are most susceptible to meningococcal disease [ 25 ]. Our patient had an unremarkable past medical history with normal complement and immunoglobulin levels with no evidence of immunosuppresion [ 26 - 28 ]. The clinical presentation of bacterial meningitis is more variable in the older as compared with the younger adult, with fewer patients manifesting with the classic symptoms of fever, neck stiffness and altered mental state than among younger adults [ 1 ]. It has been suggested that 1 of 3 findings (fever, neck stiffness, altered mental state) is present in virtually all patients with meningitis and that the absence of these features virtually excludes meningitis with a high negative predictive value (table 1 ) [ 1 ]. Our patient had none of these features on presentation and had been unwell for four weeks prior to presentation, but did develop a mild fever (<38°C) and cognitive dysfunction during her inpatient stay. The blunted febrile response is well recognised in older adults in general [ 29 ]. Our patient's CSF showed a lymphocytosis, raised protein, and low glucose ratio, which are seen in only 10% of bacterial meningitis cases. This CSF profile would normally suggest infection with Listeria monocytogenes meningitis or alternative causes such as tuberculous and fungal infection [ 30 , 31 ]. Neisseria (N) meningitidis is a leading cause of bacterial meningitis in the Western World and tends to predominate in young adults [ 19 , 20 , 25 ]. N. meningitidis is a gram-negative, aerobic diplococcus. It is classified into serogroups (e.g. A,B,C etc) according to the immunological reactivity of their polysaccharides [ 25 ]. The most prevalent serogroups implicated in clinical meningococcal meningitis are serogroup B (62%, as in our patient) and the more virulent serogroup C (22%) [ 19 , 20 ]. The relatively reduced virulence of serogroup B may partly explain the chronicity of presentation and reduced inflammatory response seen in our patient. Serogroups B and C have a seasonal variation occurring more commonly in the first quarter of the year (our patient presented in February) [ 19 ]. Meningococcal meningitis is also more common among the following groups: persons of black race; lower socioeconomic classes; those exposed actively or passively to tobacco smoke; persons exposed to overcrowding and amongst binge drinkers [ 32 - 37 ]. This case highlights the diagnostic challenge associated with bacterial meningitis presenting in an older patient. The presentation was made even more difficult owing to the blunted febrile response, the lack of inflammatory response observed in laboratory tests and the chronicity of the patient's symptoms. The diagnosis required thorough investigation during the inpatient stay. Early lumbar puncture is to be encouraged as it is essential to confirm the diagnosis. Despite a delayed diagnosis appropriate antibiotic therapy can still lead to a good outcome. Competing interests The authors declare that they have no competing interests. Authors' contributions MD generated the idea of writing the case report and was the consultant in charge of the patient. CD reviewed the case notes of the patient and wrote the original draft of the case presentation. CB significantly revised the original draft and added the conclusions, references and figures. AH offered considerable help with the manuscript revisions. All authors contributed to the final version of the manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC524513.xml
535350	Distinct gene expression profiles in different B-cell compartments in human peripheral lymphoid organs	Background There are three major B-cell compartments in peripheral lymphoid organs: the germinal center (GC), the mantle zone (MNZ) and the marginal zone (MGZ). Unique sets of B-cells reside in these compartments, and they have specific functional roles in humoral immune response. MNZ B cells are naïve cells in a quiescent state and may participate in GC reactions upon proper stimulation. The adult splenic MGZ contains mostly memory B cells and is also known to provide a rapid response to particulate antigens. The GC B-cells proliferate rapidly and undergo selection and affinity maturation. The B-cell maturational process is accompanied by changes in the expression of cell-surface and intracellular proteins and requires signals from the specialized microenvironments. Results We performed laser microdissection of the three compartments for gene expression profiling by cDNA microarray. The transcriptional program of the GC was dominated by upregulation of genes associated with proliferation and DNA repair or recombination. The MNZ and MGZ showed increased expression of genes promoting cellular quiescence. The three compartments also revealed distinct repertoires of apoptosis-associated genes, chemokines and chemokine receptors. The MNZ and GC showed upregulation of CCL20 and CCL18 respectively. The MGZ was characterized by high expression of many chemokines genes e.g. CXCL12 , CCL3 , CCL14 and IFN-associated genes, consistent with its role in rapid response to infections. A stromal signature was identified including genes associated with macrophages or with synthesis of extracellular matrix and genes that influenced lymphocyte migration and survival. Differentially expressed genes that did not belong to the above categories include the well characterized BCL6 and CD10 and many others whose function is not known. Conclusions Transcriptional profiling of B-cell compartments has identified groups of genes involved in critical molecular and cellular events that affect proliferation, survival migration, and differentiation of the cells. The gene expression study of normal B-cell compartments may additionally contribute to our understanding of the molecular abnormalities of the corresponding lymphoid tumors.	Background Appropriate T- and B-cell migration and timely interaction with antigen presenting cells (APC) are essential for the development of humoral immune responses [ 1 , 2 ]. Specialized compartments within lymphoid tissues facilitate these interactions [ 3 ]. Distinct populations of B-cells reside in these microenvironments, and, upon antigen stimulation, cells with appropriate antigen receptors differentiate and migrate among these compartments for a proper immunological reaction [ 4 - 7 ]. The initiation of a T-dependent B-cell response results from cognate interaction between a T-helper cell and a B-cell that primes the B-cell into two developmental pathways. An extrafollicular reaction takes place in the T zone and leads to the production of plasma cells with unmutated immunoglobulin (Ig) genes. The other pathway initiates a germinal center (GC) reaction, whereby activated B lymphocytes originating from extrafollicular foci enter the GC and undergo a stringent process of positive selection and affinity maturation. The selected cells differentiate into either memory B-cells or plasma cells with mutated Ig genes. The GC provides the important microenvironment for this crucial B-cell maturational process [ 8 , 9 ]. In follicles with developing GCs, the resting B-cells that are not the part of the GC response are pushed outward to form the mantle zone (MNZ) or corona around the GC B-cells. The mantle cell is a pre-GC, immunologically naïve B-cell that is also the putative cell of origin of mantle cell lymphoma [ 10 ]. These B-cells express unmutated immunoglobulin genes, sIgD high , CD27 - [ 11 ] and are mostly restricted to the follicular mantle zone [ 12 ]. In the human spleen, there is a well defined zone between the follicular B-cells and the red pulp called the marginal zone (MGZ) containing marginal-zone macrophages, granulocytes and dendritic cells that are specialized to capture blood-borne antigens and present them to the resident marginal zone B-cells [ 13 ]. Unlike primary lymphoid follicles in spleen and lymph nodes, which contain mostly mature recirculating B-cells, non-recirculating B-cells are enriched in the splenic MGZ. These cells are specially adapted to respond rapidly to T-independent (TI) antigens and have a lower threshold than recirculating or immature B-cells for activation, proliferation and differentiation into antibody-secreting cells [ 7 ]. They may therefore provide the early rapid humoral response prior to the more refined but delayed response from the GC reaction. Most adult human MGZ B-cells have the IgM high , IgD low and CD27 + phenotype, suggesting that this zone contains mainly memory B-cells [ 14 ]. Many previous studies [ 15 , 16 ] have provided important information concerning the biology of the GC. While morphology and immunophenotype are useful in defining the various B-cell compartments of peripheral lymphoid organs, the molecular signals that affect the life span, survival, retention, migration and functions of the cells in these compartments have not been widely investigated. We used the Lymphochip cDNA microarray [ 17 ] to investigate the differences in gene expression profiles in the three different B-cell compartments, the MNZ with mostly naïve B-cells, the MGZ containing memory B-cells [ 18 ] and specialized non-recirculating B-cells and the GC with a mixture of highly proliferative centroblasts and more differentiated and non-dividing centrocytes. For this study, we used both tissue compartments isolated by laser capture microdissection (LCM) and naïve and memory B-cells isolated by fluorescence-activated cell sorting (FACS). The microdissected samples contained the dominant B-cell population in each compartment as well as other cell populations in the physiological microenvironment, whereas the FACS-sorted cells contained more uniform B-cell subsets. By comparing FACS-sorted cells with the corresponding compartment from LCM, we have identified a stromal cell gene expression signature that may provide insight into stromal/B-cell interaction. Results and discussion Isolation of naïve and memory B-cells and different anatomic B-cell compartments GC and MNZ could be readily dissected from tonsillar frozen sections, but MGZ could only be reliably obtained from the spleen (Figure 1 ). Immunostaining was not applied on the sections to be microdissected because it was difficult to obtain cells from sections on charged slides and because immunostaining also led to a marked loss of amplifiable RNA from the sections, even when a rapid procedure was used [ 19 ]. Hence, immunostaining was performed on a consecutive section to guide the dissection. Immunostaining by us and others has shown that the MNZ contained over 90% B-cells, which are the IgD + CD27 -, similar to FACS-sorted naïve B-cells. The GC was easily recognizable and generally contained a higher percentage of non B-cells, including T-cells, macrophages and follicular dendritic cells (FDC). The MGZ was obtained from a spleen with a morphologically clearly defined MGZ containing mostly IgM + CD27 + B-cells, corresponding to the phenotype of FACS-sorted memory B-cells [ 14 ]. The MGZ also contained scattered T-cells and has been shown by others to contain specialized macrophages and fibroblasts [ 20 , 21 ]. The FACS-sorted populations were over 90% pure, according to post-sort immunophenotyping (Figure 2 ). Figure 1 Three different B-cell compartments isolated by LCM. Frozen sections of reactive tonsils or spleens were fixed, and a consecutive section was immunostained for CD3 to guide the dissection. The three B-cell compartments (GC, MNZ and MGZ) were isolated using LCM with the Arcturus PixCell II system (Arcturus Engineering, Mountain View, CA). Cells were captured at the 15-μm with laser set to pulse at 60 mW for 200 ms. The GC and MNZ were clearly recognizable, and the MGZ was obtained from a spleen with a morphologically clearly defined MGZ. Only well-defined GC, MNZ and MGZ were dissected to avoid contamination. Figure 2 FACS-sorting of naïve and memory B-cells from splenocytes. A single B-cell suspension prepared from a fresh spleen was isolated using the Human B-cell Isolation Kit ( See methods ) and subjected to 3-color cell sorting. Memory B-cells from the splenic B-cells were gated on the IgM high IgD low CD27 + fraction, while naïve B-cells were gated at IgM low IgD high CD27 - . FACS-sorted populations were over 90 % pure, judging from post sort immunophenotyping. Gene expression profiling analytical approach Fifteen data sets corresponding to the five sample groups were generated. Different hybridizations were correlated through a correlation matrix plot, and replicated hybridizations were shown to be closely related (R ≥ 0.85). The plots allowed us to check reproducibility of the microarray assay among different samples of each tissue (Figure 3 ). The number of genes showing differential expression between two compartments and the magnitude of difference calculated by t-statistics were further filtered by Significance Analysis of Microarrays (SAM) approach, as described previously [ 22 ]. On the Lymphochip, over 20% of the genes are represented by multiple clones, and, generally, several clones of same genes are selected by our analytical algorithm. The differentially expressed genes among the three compartments identified by SAM were grouped according to their major functional attributes and then viewed through Tree View. Figure 3 Correlation Coefficient Mapping. Reproducibility of the different hybridization experiments was checked through correlation coefficient mapping programmed in MATLAB. High correlation is seen among samples from same compartment or FACS-sorted population. Confirmation of the microarray analysis with semi-quantitative RT-PCR and with real time quantitative PCR The differential expression of some of the transcripts that had no previously reported association with any of the compartments was further validated by a semi-quantitative RT-PCR. No discrepancies were found with any of the selected genes. By PCR analysis, some of the transcripts had almost exclusive expression in one compartment: ARK2 in GC, CCL2 0 in MNZ and CMRF-35H in MGZ. Other transcripts were expressed in all compartments with a relatively high differential expression in one, such as SET and FAIM in GC, Cyclin G2 in MNZ, and NM23-H1 and CARD11 in MGZ (Figure 4 ). Figure 4 Confirmation of the Microarray analysis by semi quantitative RT-PCR. aRNA amplified from GC, MNZ and MGZ -cells was reversely transcribed and amplified by PCR. The human HPRT gene was used as the comparative standard. Five fold serially dilutions (4 dilutions) of each cDNA amplified with gene specific primers and analyzed by electrophoresis in 2% agarose gel. The transcripts had either an almost exclusive expression in one compartment ( ARK2 in GC, CCL20 in MNZ and CMRF-35H in MGZ), or they were expressed in all compartments with a significant differential expression in one – for example, SET and FAIM in GC, Cyclin G2 in MNZ and NM23-H1 , CARD11 and GAS2 in MGZ. Some of the results of the semi-quantitative RT-PCR were further validated by the SYBR ® Green real-time quantitative PCR (data not shown). The results corresponded well with both microarray and semi-quantitative RT-PCR. Gene expression characteristics in anatomic B-cell compartments Genes controlling cell proliferation and quiescence (Figure 5 , 6 ) Comparing the gene expression profiles of LCM GC and FACS-sorted GC B-cells [ 17 ] revealed that the GC B-cell signature was largely represented in the microdissected GC profile. The GC gene expression profile was dominated by the increased expression of genes associated with proliferation (e.g., CCNB1 , PCNA , Ki67 ), kinetochore association (e.g., CENPF , BUB1 and BUB3 ), functional components of mitotic checkpoints (e.g., CENPE and TTK ) and regulators of cell-cycle related events, including centrosome separation/segregation and cytokinesis (e.g. KNSL5 , ARK2 ), as expected from the known high proliferation rate of centroblasts. GC also highly expressed genes involved in DNA repair (e.g., RAD54 , BRCA2 , RAD51 , ERCC5 and MSH2 ), as expected from the frequent physiological double-strand DNA breaks associated with somatic hypermutation and isotype switching. The GC profile also showed increased levels of transcripts involved in DNA replication (e.g., DNAJ , DNA2L , DNMT1 , TOP , RFC4 and RPA1 ) and in transcription and translation (e.g., EIF2 , TAF , TCF , UHRF1 , UBD and UBE2 ). The low expression of the cyclins CCNA , CCNB1 and CCNF and of CDC2 (also known as CDK1 ) is consistent with the resting state of the MNZ and MGZ B-cells. Characteristically CCNA expression is very low in G o phase and begins to increase in early G 1 . For the cell to enter the G2/M phase, an association with CDC2 is required [ 23 ]. The transition requires CCNB1 to form a complex with CDC2 and relocate to the nucleus. This nuclear localization is mediated by CCNF [ 24 ]. However, MNZ and MGZ B-cells may also employ different mechanisms in maintaining quiescence. Cyclin G2 ( CCNG2 ) was highly expressed in MNZ cells as compared to either GC or MGZ. The function of CCNG2 differs from the conventional cyclins in negatively regulating the cell cycle [ 25 ]. Studies on HeLa cells have showed that DNA damage induces the production of cyclin G2, which then arrests the cell cycle at the G1/S boundary, and this function is independent of p53. Cyclin G2 can directly interact with the catalytic subunit of protein phosphatase 2A (PP2A) and prevent cell cycle progression. The low expression of CARD11 in MNZ may also be part of the program in maintaining the quiescent state. CARD11 has been shown to be critical for immune receptor signaling of both T and B-cells through the activation of JNK and NF-κB [ 26 ]. The increased transcriptional level of CD72 may be involved in maintaining the quiescent state in MNZ B-cells. CD72 contains an immunoreceptor tyrosine-based inhibitory motif (ITIM) in its cytoplasmic domain and functions as a negative regulator of B-cell signaling [ 27 ]. Interestingly, many genes associated with proliferation were expressed at even lower levels in MGZ than in MNZ cells. These cells highly expressed growth inhibitory genes such as CMRF-35H [ 28 ], CBL-B [ 29 ], and GAS2 [ 30 ], which may contribute to the quiescent state in MGZ B-cells. Figure 5 Cell proliferation and quiescence. Differential gene expression among B-cell compartments. Genes identified in SAM analyses were merged according to functional or operational categories and visualized in Tree View. Color changes within a row indicate expression levels relative to the median of the sample population. Only transcripts differing 2-fold or more in their magnitude than the median or mean of the other two compartments are shown. Apoptosis (Figure 7A ) The markedly decreased BCL2 expression in GC B-cells makes them vulnerable to undergo apoptosis unless rescued by survival signals [ 31 ]. An increase in the expression of proapoptotic genes e.g. BIK , FASL and PDCD8 ( AIF ) suggests a further increase in susceptibility to apoptosis in the GC. However, FAIM is overexpressed in the GC and may represent a protective mechanism in GC B-cells that have appropriate BCR signaling and CD40L stimulation with resultant upregulation of FAIM and increased resistance to FASL-mediated apoptosis [ 32 ]. Presumably, GC B-cells with sIg having poor antigen affinity will be ineffective in activating FAIM. In addition, a TNF receptor family member (TNFRSF17/BCMA) which promotes the B-cell survival [ 33 ] showed increased transcription in the GC. Programmed Cell Death 4 (PDCD4), which functions mainly as an inhibitor of translation by inhibiting the activity of eIF4A helicase, which helps to unwind the 5' end of mRNAs, was markedly repressed in the GC B-cells [ 34 ]. This suggests that PDCD4 repression facilitates the rapid proliferation of centroblasts, which requires a high rate of protein synthesis. Both MNZ and MGZ had higher expression of BCL2 , but have different profiles of other apoptosis/survival genes that may represent specific adaptation of these cells to their unique physiological states and microenvironment. The expression of BNIP3 , encoding a proapoptotic protein of the BCL2 family[ 35 ], is markedly down regulated in MGZ cells, perhaps providing additional protection against apoptosis in memory B-cells. On the other hand, TCL1 was upregulated in MNZ only and may have an antiapoptotic role in that population. There was increased expression of Suppressor of Death Domains (SODD) in MGZ cells, suggesting a complex regulation of signaling through the TNFR superfamily. SODD is associated with TNFR1 in vivo, maintaining the receptor in an inactive monomeric state. The release of SODD from TNFR1 permits the recruitment of proteins such as TRADD and TRAF2 to the activated TNFR1 signaling complex [ 36 ]. It has been demonstrated that TNF-induced activation of NF-κB is accelerated in SODD-deficient cells. The high expression of SODD may be a major mechanism to dampen TNFR1 signaling in MGZ B-cells in the resting state. The higher expression of CARD11 in MGZ may have a pro-survival function, but it may also have a role in MGZ organization. It has been shown that loss of CARD11 in mice resulted in the complete loss of CD5 + peritoneal cells and reduced number of IgD high IgM low mature splenic B-cells, indicating its role in B-cell development [ 37 ]. Two closely related genes, NM23-H1 and NM23-H2 , which share an amino acid identity of 88%, were highly expressed in MGZ. NM23 H1 is a granzyme A-activated DNase (GAAD) that is inhibited by SET [ 38 ]. The high expression of NM23-H1 and the low expression of its inhibitor SET was opposite in their expression profile in the GC, suggesting that this expression may influence apoptosis in opposite directions in these two B-cell compartments. Figure 6 DNA repair, replication and protein synthesis. Figure 7 (A) Apoptosis and cell survival and (B) Cytokines and chemokines and their receptors. Chemokines, cytokines and their receptors (Figure 7B ) Chemokines attract primary B-cells and play an important role in the homing and localization of B-cell subsets at different stages of antigen-independent and dependent reaction [ 39 ]. Our microarray data revealed that CCL18 , encoding a chemokine secreted by immature dendritic cells (DC), is specifically upregulated in the GC compartment. Our finding was supported by a recent report showing that CCL18 is produced by GC DC and can attract MNZ B-cells towards GC [ 40 ]. The higher expression of CCL18 may be especially important during the onset of a GC reaction, the time point to recruit antigen primed pre-GC B-cells, which then interact with GC DC to initiate and maintain the GC reaction. The GC compartment showed increased expression of CXCL10 ( IP-10 ). which has pleiotropic effects, including stimulation of monocytes and T cell migration [ 41 ]. The GC also showed increased transcriptional level of genes that may suppress or control inflammatory responses; e.g., SOCS1 limits cellular response to IFNγ, IL-2 and IL-6[ 42 , 43 ]. Macrophage inhibiting cytokine 1 ( MIC1 , also known as PLAB ) [ 44 ], is only expressed by activated macrophages, but not by resting macrophages. Its higher expression in the GC may reflect the presence of moderate numbers of macrophages and its possible role in suppressing the inflammatory response in the GC. Increased expression of the chemokine CCL20 was observed in MNZ cells. Human naïve and memory B-cells express the cognate receptor for CCL20, CC-chemokine receptor 6 (CCR6) [ 45 ]. The high expression of CCL20 may play a vital role in naïve B-cell migration and localization in the MNZ. The chemokine gene CXCL12 (also know as stromal cell-derived factor 1, SDF1 ) is highly expressed in MGZ cells. The receptor for CXCL12 is CXCR4, which is present on CD34 + cells, myeloid cells, CD4 + T cells, B-cells, epithelial cells, endothelial cells, and dendritic cells. In the bone marrow, stromal cells secrete CXCL12, which is involved in the emigration of hematopoietic precursors to the marrow during embryogenesis [ 46 ]. In peripheral lymphoid organs CXCL12 may be involved in the migration of B-cells and possibly other cells, such as T cells and plasma cells, to the MGZ. CCL14 (also known as HCC1 ) and CCL3 (also known as MIP-Iα ) were more highly expressed in the MGZ. CCL14 can activate human monocytes via receptors that also recognize CCL3 [ 47 ]. CCL3 is a proinflammatory cytokine important in the clearance of viral infections and the stimulation of the innate immune response [ 48 ]. Thus, the expression of this gene may be important in innate immunity in the MGZ of the spleen. CXCL13 and CCL5 were upregulated in both microdissected MGZ and MNZ compared with the FACS-sorted B-cell populations. Previous studies have established an important role for CXCL13 (BLC) in the development of Peyer's patches (PP) and many peripheral lymph nodes. It also controls B-cell migration and thus the organization of B-cell areas [ 49 ]. CCL5 (RANTES), a stromal related chemokine, elaborated by activated T, NK and macrophages has been shown to interact with CD44 to activate the MAPK pathway [ 50 ]. It is possible that CCL5, under appropriate conditions, contributes to cellular activation that may be particularly relevant to MGZ B-cells, in which rapid response on recognition of antigen stress signal is important. A number of IFN-induced genes ( AIM2, IFNGR1 , and IFNAR -1 and - 2 ) were also preferentially expressed and may reflect the unique function of the MGZ to provide the first rapid response to particulate or T cell-independent antigens. The MGZ also showed increased expression of many members of G protein pathways consistent with more active chemotaxis, cell motility and secretory functions. In addition, MGZ cells showed higher expression of IL-7Rα , consistent with the role of the IL-7R in MGZ organization [ 51 ]. Aside from its role in B-cell differentiation and proliferation IL-7Rα expression is also required for the recruitment of precursor cells to develop in secondary lymphoid organs and for the proper structural organization of these organs [ 52 ]. Figure 8 Stromal signature. Extracellular matrix and stromal signature (Figure 8 ) Cells function within the context of a three-dimensional (3D) extracellular matrix (ECM) that participates in regulating cellular motility, proliferation and survival. In the GC, COL9A3 , which encodes collagen IX [ 53 ], and COL2A1 , which encodes collagen XI [ 54 ], were uniquely overexpressed. In the marginal zone COL14A1 , COL16A1 , COL3A1 and COL6A3 were expressed at higher level, which suggests a role for these genes in the synthesis of specific extracellular matrix. There was also a marked overexpression of macrophage metalloelastase 12 ( MMP12 ), encoding a metalloproteinase that preferentially degrades elastin and takes part in the remodeling of extracellular matrix. No collagen-specific gene was up-regulated in the MNZ. Microdissected compartments contained a minor component of stromal T cells, macrophages, dendritic cells and fibroblasts whereas FACS-sorted cells from lymphoid tissues comprise almost exclusively B-cells. Thus, an insight into the gene expression profile of the stromal elements can be obtained by comparing the expression profile of FACS-sorted and microdissected cells. We found a set of genes that likely represent the stromal signature. Osteonectin (SPARC) , upregulated in LCM samples, encodes a matrix-associated protein that elicits changes in cell shape, inhibits cell-cycle progression, and influences the synthesis of extracellular matrix [ 55 ]. It regulates endothelial barrier function through F-actin-dependent changes in cell shape [ 56 ]. Two members of the Maf family ( MafB , and c-Maf ) were also part of the stromal signature. The Maf family of genes encode bZip nuclear transcription factors and play an important role in morphogenesis and cellular differentiation [ 57 ]. These genes are expressed in a variety of organs, including the spleen, in agreement with our finding. The MGZ expressed elevated levels of ICAM1 and VCAM1 . MGZ B cells express the integrin LFA1 which binds to its ligands ICAM1 and VCAM1, and this interaction may control the localization of these B cells [ 58 ] in this compartment. Our results also showed elevated expression of VCAM1 , ITGAL (LFA-1) and ITGA6 in the MNZ, suggesting a role for these adhesion molecules in mantle cell localization as well. The kruppel-like transcription factor BCL11a (also called Evi9), which is essential for normal B-cell lymphopoiesis, was upregulated in LCM cells only. Interestingly, bone marrow from BCL11a -/- mouse can induce thymic lymphoma in wild type mice. Thus, the increased expression of BCL11a in the MNZ and MGZ may be physiologically relevant to the function of lymphocytes in these regions [ 59 ]. Figure 9 Other unique compartment markers. Other differentially expressed genes (Figure 9 ) Many genes know to be specifically expressed in GC B-cells are found to be upregulated: e. g., BCL6 , CD10 , GCET1 , GCET2 , JAW1 and CD38 . A number of genes were clearly upregulated in the MNZ or MGZ but their functional significance is largely unknown. Some of these would be interesting targets for further investigation. Among the genes encoding surface molecules, CD59 was highly expressed in the GC. CD59 antigen is a small protein that inhibits complement-mediated pore formation or lysis by preventing the formation of membrane-embedded C9 multimers [ 60 ]. It is likely that the over expression of CD59 in GC can prevent complement-mediated damage to FDCs with entrapped immune complex. CD10 and CD38 are well established markers of GC B-cell and over expression of the corresponding mRNA in the GC is expected [ 61 - 63 ]. Notably, CIITA was markedly down regulated in GC cells, associated with a general low expression of MHC transcripts. Conclusions The gene expression profiles of the three B-cell compartments reflect distinctive functional attributes of the resident B-cell populations. They also showed different molecular microenvironments that allow the different B-cell populations to differentiate and function properly. GC B-cells have a high proliferation gene signature, whereas MNZ and MGZ cells are characterized by signals that help to maintain the quiescent state. Genes involved in the apoptosis pathway are differentially expressed in the three B-cell compartments, reflecting different adaptations for survival in different B-cell populations. Expression of different chemokines, their receptors, and stromal molecules have been detected. Many of these have been implicated in the establishment of the normal lymphoid architecture in peripheral lymphoid organs and in attracting distinct immune-cell populations to specific lymphoid areas. The expression of unique sets of genes may also reflect the functional adaptation of cells in a specific location, such as genes involved in DNA repair in the GC and genes that are active in innate immune response to infection in the MGZ. Gene expression profiling of B-cell compartments has allowed us to obtain a global survey of the molecular signals that are functionally important in B-cell subpopulations as well as the respective microenvironments. One of the major challenges is to delineate the functions of the uncharacterized genes that are unique to each of the compartments. Another challenge is to exploit these normal transcriptional profiles to further our understanding of the normal immune response and the derangements resulting in the corresponding lymphoid tumors. Methods Laser capture microdissection (LCM) Tissue blocks of tonsils and spleens were snap frozen in O.C.T immediately after surgery. Four micrometer thick frozen sections of reactive tonsils or spleens on plain glass slides were fixed with 70% ethanol for 30 seconds, rinsed in DEPC water and stained with hematoxylin for 30 seconds, followed by another water rinse. The sections were then dehydrated with 70%, 95% and 100% ethanol for 10 seconds each. Finally, the slides were passed through xylene twice, each for 30 seconds. A consecutive section was immunostained for CD3 to guide the dissection. The three B-cell compartments (GC, MNZ and MGZ) were isolated using LCM with the Arcturus PixCell II system (Arcturus Engineering, Mountain View, CA). To avoid contamination, only well-defined GC, MNZ and MGZ were dissected. Cells were captured at the 15-μm laser setting on CapSure LCM Caps (Arcturus). The laser was set to pulse at 60 mW for 200 ms. The Institutional Review Board of the University of Nebraska approved the usage of tissues for this study. Cell preparation and FACS sorting Tissue from fresh spleens or tonsils was cut into small pieces in cold RPMI-1640, and cells released by grinding with a glass tissue homogenizer. The crude cell suspension was passed through a nylon mesh (Spectrum Laboratories, Inc) to generate a single-cell suspension. B-cells were firstly isolated using the Human B-cell Isolation Kit (a cocktail of hapten-modified antibodies to CD2, IgE, CD4, CD11b, CD16 and CD36) and the Midi Macs system (Miltenyi Biotec, Auburn, CA). The highly enriched B-cell population (negative fraction) was subjected to 3-color cell sorting. Briefly, 1 × 10 7 B-cells were stained with IgM-Cy-chrome, IgD-FITC and CD27-PE (BD Pharmingen, SanDiego, CA) at 4°C for 30 min. MGZ B-cells were isolated from the splenic B-cells gated on the IgM high IgD low CD27 + fraction, whereas MNZ B-cells were selected based on IgM low IgD high CD27 - using the BD FACSVantage™ SE high-speed cell sorter (Becton-Dickinson, SanJose, CA) RNA extraction and T7 RNA amplification Total RNA was extracted from each sample of microdissected cells with Trizol™ (Gibco BRL, Carlsbad, CA) and further purified with the RNeasy Mini Kit (Qiagen, Valencia, CA). RNA amplification was performed using a modified Eberwine protocol [ 64 ]. Briefly, first-strand cDNA was synthesized by reverse transcription using oligo dT(25)-T7 anchoring primer and superscript II at 42°C for 1 hour. Second-strand synthesis was performed with 40 units E. coli DNA polymerase I, 2 units RNase H, 10 units E. coli DNA ligase (Life Technologies, Carlsbad, CA) in 150 μl volume. Antisense RNA (aRNA) was generated by in vitro transcription (IVT) using the Ampliscribe™ High yield Transcription Kit (Epicentre Technologies, Madison, WI) containing 1000 units AmpliScribe T7 enzyme at 37°C for 8–12 hours, as per the manufacture's instruction. Second-round amplification and IVT were performed as described previously [ 65 ]. The quality and quantity of aRNA were monitored on agarose gel electrophoresis and by spectrophotometer. Typically, 30–50 μg of aRNA were generated from each 10 ng of total RNA by two rounds of amplification. Gene expression profiling using the Lymphochip Analysis of gene expression was performed using the Lymphochip cDNA microarray, which contained 15,132 cDNA clones representing 7399 known or uncharacterized genes [ 66 ]. Labeled cDNA from each compartment was first hybridized with a test cDNA microarray to assess the quality and quantity of the amplified aRNA before using them on the Lymphochip. In each experiment, reverse transcription was carried out on 8–9 μg of aRNA, and aminoallyl-dUTP was incorporated into the cDNA using a dUTP: dTTP ratio of 4:1 . The aminoallyl group on the dUTP reacts with the ester group on the cyanine dyes. Cy3 dye was used to label the standard cDNA and Cy5 dye the test probe, and hybridization was performed as previously described [ 17 ]. Data and statistical analysis procedure Each tissue type was independently isolated, amplified and profiled in three separate experiments to enhance the reliability of the gene expression data. Images of hybridized microarrays were obtained and processed using GenePix 4000B microarray scanner (Axon Instruments, Union City, CA). Spots or areas of an array with obvious blemishes were flagged and excluded from subsequent analyses. Fluorescence ratios were normalized for each array by applying a single scaling factor to all fluorescent ratios from the array [ 17 ]. The correlation coefficients among 15 hybridized cDNA microarrays were calculated and expressed in Correlation Coefficients Mapping (CCM), programmed in MATLAB © (Mathworks, Inc, Natick, MA), which provided an overview of the similarity of expression profiles between multiple samples. Only genes with at least two values out of the triplicate experiments showing similar behavior were included for analysis. The expression data for each gene from an individual compartment was median/mean centered with weighted variance across the two or three replicates showing similar behavior. The initial data reduction was performed using the two-tailed student t-test to compare the differences in gene expression levels between individual compartments. Genes differentially expressed between the two compartments with a p-value of less than 0.05 were selected for further analysis using the Significance Analysis of Microarrays (SAM) approach, as described previously [ 22 ]. SAM assigns each gene a score based on its change in average expression between two groups, relative to the gene's standard deviation of permutated measurements. The scatter plots for observed relative difference vs expected relative difference between two compartments were used to find the potentially significant genes and plotted in the T-distance histogram correlating with the p-values. The genes selected from the common set of the analysis result from both t-statistics and SAM were grouped according to their functional characteristics after analyzing through OMIM or Gene Ontology database ( or ) and viewed by TreeView. Semi-quantitative and real-time quantitative PCR To confirm the differential mRNA expression of the genes identified by the Lymphochip in different B-cell compartments, a semi-quantitative RT-PCR was employed. In brief, 200 ng aRNA was reversely transcribed into cDNA with 200 ng random primer using MMLV-RNase H - reverse transcriptase as per the manufacturer's instructions (Invitrogen, Carlsbad, CA). Five-fold serially diluted cDNAs from GC, MNB and MZB were amplified with gene-specific primers for 30 cycles with the following cycling conditions: A denaturation step at 94°C for 2.5 minutes and then each PCR cycle at 94°C for15 sec, 52°C for 30 sec, and 72°C for 30 sec followed by a final extension at 72°C for 5 min. The human HPRT transcript was used as the comparative standard. The products were analyzed by electrophoresis in 2% agarose gel. The primers were designed to amplify the cDNA close to the 3' end of the transcript, and all the PCR products were less than 200 bp in length. Some of the results from the semi-quantitative RT-PCR were also validated by the real-time quantitative PCR with DyNAmo™ HS SYBR ® Green qPCR Kit (MJ Research, Reno, NV) on DNA Engine OPTICON2 (MJ Research, Reno, NV) as per the manufacturer's instructions. The PCR protocol used an initial denaturation of 95°C for 15 minutes followed by 35 cycles (95°C for 10 sec, 52°C for 15 sec and 70°C for 20 sec). The plate was read at 70°C according to the melting point of the amplicon. Serial dilutions of cDNA from the lymphoid standard [ 67 ] were used to construct standard curves for the target genes ( FAIM , CCL3 , SODD , NM23-H1 , CARD11 , Cyclin G2 and CIITA-8 ) and the endogenous reference genes (HPRT). For each unknown sample, the relative amounts of target cDNAs and reference cDNAs applied to the PCR reaction system were calculated using linear regression analysis from the corresponding standard curves [ 68 ]. Then the normalized expression level of the target gene in each sample was calculated by dividing the quantity of the target transcript with the quantity of corresponding reference transcript. The normalized values of the target transcript were used to compare its relative expression levels in different samples. Authors' contributions YS carried out the LCM, in vitro RNA amplification and semi-quantitative PCR. JI participated in the design of the study, microarray procedure, data analysis, and drafted the manuscript. LX, RL and SS participated in data analysis. JE provided the microarray facility and various technical assistance and advice. LS and AR provided the reference standard, the Lymphochip and helpful discussions. KD, GZ, TM participated in helpful discussions and interpretation of the data. WCC conceived, organized and supervised the study, and participated in the analysis and interpretation of the data. All authors have read and approved the final manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC535350.xml
554786	Using equilibrium frequencies in models of sequence evolution	Background The f factor is a new parameter for accommodating the influence of both the starting and ending states in the rate matrices of "generalized weighted frequencies" (+gwF) models for sequence evolution. In this study, we derive an expected value for f , starting from a nearly neutral model of weak selection, and then assess the biological interpretation of this factor with evolutionary simulations. Results An expected value of f = 0.5 (i.e., equal dependency on the starting and ending states) is derived for sequences that are evolving under the nearly neutral model of this study. However, this expectation is sensitive to violations of its underlying assumptions as illustrated with the evolutionary simulations. Conclusion This study illustrates how selection, drift, and mutation at the population level can be linked to the rate matrices of models for sequence evolution to derive an expected value of f . However, as f is affected by a number of factors that limit its biological interpretation, this factor should normally be estimated as a free parameter rather than fixed a priori in a +gwF analysis.	Background Felsenstein [ 1 ] was the first to introduce an evolutionary model for DNA sequences, which allows for unequal nucleotide frequencies (see also [ 2 ]). His F81 model allows for substitutions at a rate proportional to the frequencies of the ending nucleotides. It is considered the simplest rate matrix for accommodating variable nucleotide frequencies and is therefore the starting point for the consideration of more complex models with frequency variation (e.g., the HKY model of Hasegawa et al . [ 3 ]). Goldman and Whelan [ 4 ] described new variants of these F81-based models (their +gwF (generalized weighted frequencies) models; e.g., JC+gwF for Jukes and Cantor [ 5 ], and K2P+gwF for Kimura [ 6 ]). At the heart of their +gwF variants was a new free parameter ( f ) to accommodate the frequencies of the starting, as well as ending, nucleotides in the evolutionary process: where q ij refers to the substitution rate from nucleotide i to j , π i and π j correspond to their equilibrium base frequencies, and s ij is the exchangeability between the two. In the +gwF variants, the substitution rate becomes more dependent on the ending nucleotide as f decreases from 1 to 0, with f = 0 for the classic F81-type models. This study starts with a population genetics model to derive equations that link weak selection, genetic drift, and mutation to the f parameter and evolutionary rate matrices of the +gwF variants. These theoretical derivations lead to an expected value of f = 0.5. However, as illustrated with simulations, the f parameter is complex and thus its biological interpretation must be considered with caution. Results Derivation of the rate matrix for the weak selection model The nearly neutral model of molecular evolution states that most DNA mutations of longer-term evolutionary consequence are under weak selection and are therefore prone to drift [ 7 , 8 ]. For a diploid population of size N , a neutral mutation has a probability of 1/2 N of becoming fixed in the population. However, because of drift, even slightly deleterious mutations can become fixed, but at a probability of less than 1/2 N . Advantageous mutations have higher fixation probabilities than neutral mutations. In the nearly neutral model, the distribution of alleles is determined by an equilibrium of selection, drift, and mutation. Consider a number of sites under identical evolutionary constraints and with a bias in nucleotide distribution. Assume that weak selection and drift are the causes of this bias; e.g., as for the codon usage biases in micro-organisms and Drosophila [ 9 , 10 ]. In our model, some nucleotides confer a slightly higher fitness onto the organism than do others, regardless of their position, and these can become fixed in the population through drift and/or selection. Here, we also assume that selective advantages are additive for the two alleles of the diploid organism [ 11 , 12 ]. Let the selective advantages of the four nucleotides be given by s A , s C , s G , and s T . The differences between these selection coefficients will be very close to zero, since no strong selection is expected. Consider a mutation from nucleotide i to j , with a selective advantage of s = s j - s i (a selective disadvantage exists when s is negative). For a population of size N and an effective size of N e , Kimura [ 11 ] showed that the fixation probability in this population is given by: when s ≠ 0. For s = 0, we have P ( s ) = 1/2 N . This approximation is valid for small values of s , which is the case here. The substitution rate from nucleotide i to j is proportional to P ( s j - s i ): q ij = 2 N μ ij P ( s j - s i ), (3) where μ ij is the mutation rate from i to j . For different i and j , μ ij can vary because of unequal transition versus transversion rates (for example). Furthermore, let us assume that the mutation rate is the same for either direction of substitutions between i and j . This assumption is necessary to maintain the widely used condition of time reversibility in the evolutionary process, which thereby keeps the following derivations tractable [ 1 , 13 ]. We then have: Since q ij / q ji can be written as a function evaluated at s j divided by the same function evaluated at s i , evolution is time reversible according to this model with: Here, c and c ' are constants with c ' = -l/4 N e log c , which will be chosen to make the equilibrium frequencies sum to one. The substitution rates can now be approximated as: Given an exchangeability of s ij = μ ij , this equation reduces to equation (1) with f = 0.5 and an adjustment factor of: This adjustment factor is close to one for moderate ratios of π , with a horizontal tangent around π j / π i = 1 and a slight bending downwards when deviating from this value (Fig. 1 ). Thus, a value of f = 0.5 is suggested for the +gwF variants according to these derivations of the weak selection model. Figure 1 Adjustment factor as a function of the ratio of π 's . The adjustment factor is given by (equation (7)). Evolutionary simulations Evolutionary simulations were conducted to examine the effects of violating certain assumptions in the above model of weak selection. Unless otherwise noted, these simulations were based on the K2P+gwF model with f = 0.5 and k = 2 (for the transition/transversion ratio). Simulations consisted of four sequences of length 10,000 and relied on a symmetric rooted phylogeny with all branch lengths equal to 0.10 expected substitutions per site under the model in question [i.e., ((seq1:0.10, seq2:0.10):0.10, (seq3:0.10, seq4:0.10):0.10)]. Violations of the weak selection model were incorporated in the simulations by: (1) heterogeneous sequences with sites drawn from different equilibrium base frequencies; (2) populations in disequilibrium due to changing N e ; and (3) an accelerated C to T substitution rate. Estimates of f for the simulated sequences were made with the K2P+gwF model. Forty simulations were run for each test condition, with the results for the f estimates summarized as their means and twice their standard errors. In the first set of simulations, six categories of sites with different equilibrium distributions were considered (Table 1 ). The f estimates for the simulations with each category alone were not significantly different from 0.5 (i.e., the value under which the sequences were generated). In contrast, for the simulated heterogeneous sequences (i.e., those composed of equal numbers of sites from two or three different categories), their values of f varied significantly in either direction from 0.5. Analogous results were obtained for the simulations of homogeneous and heterogeneous sequences under the HKY model (with f = 0.0 instead of 0.5). Thus, the value of f can vary considerably when heterogeneous sequences are analyzed with a +gwF model. Here, such deviations are a consequence of using a single rate matrix to analyze sequences that were derived from two or three different ones. Table 1 Starting equilibrium base frequencies and results for the simulations with either homogeneous or heterogeneous sequences (i.e., those with sites from single versus multiple categories, respectively). Categories π C π G π T Bias b f c A 0.10 0.40 0.30 0.20 0.154 0.50 ± 0.01 0.00 ± 0.01 B 0.30 0.30 0.30 0.10 0.105 0.50 ± 0.01 0.00 ± 0.01 C 0.30 0.20 0.20 0.30 0.029 0.50 ± 0.02 0.00 ± 0.03 D 0.40 0.20 0.20 0.20 0.078 0.49 ± 0.01 0.01 ± 0.01 E 0.20 0.40 0.20 0.20 0.078 0.51 ± 0.01 -0.01 ± 0.02 F 0.20 0.20 0.40 0.20 0.078 0.50 ± 0.02 -0.01 ± 0.01 A+B 0.20 0.35 0.30 0.15 0.074 0.43 ± 0.01 -0.11 ± 0.02 A+C 0.20 0.30 0.25 0.25 0.015 0.34 ± 0.03 -0.16 ± 0.03 B+C 0.30 0.25 0.25 0.20 0.015 0.24 ± 0.02 -0.33 ± 0.03 A+B+C e 0.23 0.30 0.27 0.20 0.016 0.39 ± 0.04 -0.13 ± 0.04 D+E+F e 0.27 0.27 0.27 0.20 0.010 0.68 ± 0.03 0.29 ± 0.04 a Expected nucleotide distribution. b Nucleotide bias, as information content measured in bits: . c Mean ± twice the standard error of the estimate. d f = 0:0 for these simulations with the HKY model. With f = 0:0, the HKY+gwF variant is reduced in these simulations to its more standard F81 based model. e The heterogeneous sequences in these simulations were of length 9,999, rather than 10,000, since the latter is not a multiple of 3. In the second set of simulations, N e was kept constant until the time of the most recent common ancestor for the four simulated sequences. Then, N e was either left unchanged or was suddenly changed by a certain factor. The latter was done by replacing the rate matrix derived from equation (4), resulting in new equilibrium frequencies of the nucleotides. When N e was kept constant, the selective pressures and drift were left unchanged, thereby maintaining the same starting equilibrium frequencies throughout the phylogeny. Thus, the corresponding f estimates did not significantly differ from 0.5 (Fig. 2 ). In contrast, increases in N e lowered the value of f as the efficiency of selection was increased relative to drift [ 4 ]. Correspondingly, the evolutionary process became more dominated by the ending nucleotide. This increasing dominance can be expected to continue until a new equilibrium is restored (which occurs on a longer time scale than that in these simulations). Figure 2 Two situations where f is affected by deviations from the model . (A) The effect of a change in N e on the value of f . This change in N e occurs in the most recent common ancestor of the four simulated sequences. Population ratio refers to its N e after versus before this change. (B) The effect of an increased C to T substitution rate. Categories A, B, and C are defined in Table 1. In the third set of simulations, an acceleration in the C to T substitution rate was incorporated, thereby modeling an increase in their mutation rate due to the deamination of methylated C's in CpG pairs [ 14 ]. The introduction of this bias resulted in significant deviations of f in either direction from 0.5, even though their sequences were simulated in equilibrium (Fig. 2 ). Thus, the value of f can vary considerably when the rates for reciprocal mutations are unequal. Discussion This study illustrates how selection, drift, and mutation within a population can be linked to the f parameter and rate matrices of the +gwF variants for sequence evolution. Our weak selection model relies on the fixation probabilities of mutant alleles with additive genie selection and equal mutation rates for reciprocal substitutions. What is now needed are additional studies that link other population genetics models to the +gwF variants [ 9 ]. For example, the population genetics models of Li [ 15 ], which focus on allele frequency distributions and different modes of selection and mutation, could be studied for their connections to the f parameter and +gwF rate matrices. Collectively, the three sets of simulations highlight that the f parameter is complex and can be influenced by a number of different factors [ 4 ]. This complexity limits its biological interpretation and the use of its expected value of 0.5 as derived for the weak selection model. Correspondingly, in many +gwF analyses, f ill need to be estimated as a free parameter rather than fixed beforehand. Goldman and Whelan [ 4 ] focused on amino acid sequences, where they found that the +gwF models provided better fits to the majority of their protein data sets. They also analyzed two rather small nucleotide data sets for which the general reversible model (REV) outperformed the +gwF variants. As noted by them, the REV model provides enough free parameters to cover the effects of a +gwF analysis. Thus, given sufficient data, this model will consistently outperform the simpler +gwF variants, since it can always accommodate more of the evolutionary process by virtue of its extra parameters. Nevertheless, as widely acknowledged, simpler models have their place, since they allow one to maximize analytical power for more limited data, while minimizing the risk of over-parameterization [ 13 , 16 ]. Thus, as for the JC, K2P, and HKY models, we expect their +gwF variants to remain of interest as part of the hierarchy of simple to complex models for sequence evolution. Authors' contributions Both authors contributed to the conception and design of this study and to the writing, reviewing, and final approval of this article. B.K. performed the simulations and parameter estimations.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC554786.xml
544595	The migration of physicians from sub-Saharan Africa to the United States of America: measures of the African brain drain	Background The objective of this paper is to describe the numbers, characteristics, and trends in the migration to the United States of physicians trained in sub-Saharan Africa. Methods We used the American Medical Association 2002 Masterfile to identify and describe physicians who received their medical training in sub-Saharan Africa and are currently practicing in the USA. Results More than 23% of America's 771 491 physicians received their medical training outside the USA, the majority (64%) in low-income or lower middle-income countries. A total of 5334 physicians from sub-Saharan Africa are in that group, a number that represents more than 6% of the physicians practicing in sub-Saharan Africa now. Nearly 86% of these Africans practicing in the USA originate from only three countries: Nigeria, South Africa and Ghana. Furthermore, 79% were trained at only 10 medical schools. Conclusions Physician migration from poor countries to rich ones contributes to worldwide health workforce imbalances that may be detrimental to the health systems of source countries. The migration of over 5000 doctors from sub-Saharan Africa to the USA has had a significantly negative effect on the doctor-to-population ratio of Africa. The finding that the bulk of migration occurs from only a few countries and medical schools suggests policy interventions in only a few locations could be effective in stemming the brain drain.	Background Doctors migrate from developing countries to wealthier countries in order to further their careers, or improve their economic or social situation. The World Health Organization (WHO) has long recognized that migration of health personnel from developing to developed countries creates unfortunate imbalances in the global health workforce [ 1 ]. America's physician workforce has been significantly infused with foreign-trained international medical graduates (IMGs) since World War II. The purpose of this paper is to describe a sub-population of IMGs in the USA, those who have trained in one of the 47 African subcontinent nations. African governments have been very clear about their objections to the wholesale migration of their physicians to rich countries. In 1996, South Africa's then-Deputy President Thabo Mbeki implored the World Health Assembly to take measures to stop the flow of physicians from poor countries to rich ones. In 1995, South Africa itself banned the recruitment of doctors from other Organization of African Unity countries [ 2 ]. Nonetheless, large numbers of African-trained physicians leave home upon completion of their medical school training in search of careers in higher-income countries. They leave behind health systems in sub-Saharan Africa that are severely stressed: life expectancy is only 50 years, 162 children in 1000 die before reaching their fifth birthdays, and only half have access to clean water sources [ 3 ]. Further, AIDS prevalence among those 15 to 49 years old is estimated to be 8.4% [ 4 ], and in four countries, adult HIV prevalence exceeds 30% [ 5 ]. While health improvements in Africa will require a broad agenda of development activities, access to an educated workforce of health professionals is also essential [ 6 ]. African country health systems and workforce data are poor, making it difficult to estimate the effects of physician migration on sending countries. The World Bank has documented this data gap, noting "Quantitative data on the health workforce is notoriously unreliable in most countries...In poor countries, government and professional information systems are weak, when they exist at all, and are rarely comprehensive (often there is no information on the private sector) and up-to-date" [ 7 ]. Indeed, the way many African country ministries of health learn about the extent of their own emigration is through gleaning data presented by destination countries [ 8 ]. This paucity of sending-country data makes it difficult to fully describe the impact of migration on countries of origin. The 47 nations of sub-Saharan Africa have a total of 87 medical schools, although 11 countries have no medical school at all and 24 have only one each (see Table 1 ). The population of the subcontinent totals over 660 million people, with a ratio of fewer than 13 physicians per 100 000 population, or a total of 82 949 doctors [ 9 ]. By comparison, the United Kingdom (UK) has 164 physicians per 100 000 and the USA has over 279 physicians per 100 000 (or almost 800 000 doctors for a population of 284 million). Table 1 Physician workforce distribution and number of medical schools by African country Country Population (in 1000s) Physicians per 100 000 population Total number of physicians FAIMER number of medical schools Angola 10 132 7.7 780 1 Benin 6428 5.7 366 1 Botswana 1578 23.8 376 0 Burkina Faso 12 217 3.4 415 1 Burundi* 5714 6 343 1 Cameroon 14 792 7.4 1095 1 Cape Verde 0.04 17.1 68 0 Central African Republic 3501 3.5 123 1 Chad 8419 3.3 278 1 Comoros 0.578 7.4 43 0 Congo 2809 25.1 705 1 Congo (DR) 51 810 6.9 3575 3 Côte d'Ivoire 15 866 9 1428 1 Equatorial Guinea 0.474 24.6 117 0 Eritrea 4232 3 127 0 Ethiopia* 62 651 2 1253 3 Gabon* 1223 20 245 1 Ghana 19 509 6.2 1210 3 Guinea 8642 13 1123 1 Guinea-Bissau 1278 16.6 212 1 Kenya 30 310 13.2 4001 2 Lesotho 1847 5.4 100 0 Liberia 3149 2.3 72 2 Madagascar 15 506 10.7 1659 3 Malawi* 10 874 2.3 250 1 Mali 10 665 4.7 501 1 Mauritania 2668 13.8 368 0 Mauritius 1179 85 1002 1 Mozambique# 16 934 2.57 435 1 Namibia 11 826 29.5 3489 0 Niger 10 174 3.5 356 1 Nigeria 123 750 18.5 22 894 16 Rwanda* 7405 4 296 1 Sao Tome & Principe 0.16 46.7 75 0 Senegal 9784 7.5 734 2 Seychelles 0.08 132.4 106 1 Sierra Leone 5203 7.3 380 1 So. Africa 42 351 56.3 23 844 8 Somalia 7253 4 290 1 Sudan 35 080 9 3157 14 Swaziland 1,120 15.1 169 0 Tanzania 33 768 4.1 1384 4 The Gambia 1367 3.5 48 0 Togo 5033 7.6 383 1 Uganda* 23 496 3 705 3 Zambia 9799 6.9 676 1 Zimbabwe 12 186 13.9 1694 1 TOTAL/AVG 663 529 12.5 82,949 1.8 avg Avg. physicians per country 1765 Indicates countries for which physicians/100 000 data came from World Bank instead of World Health Organization # Mozambique data come from Stephen S. Gloyd, as no data are available from WHO or World Bank Note on calculations: Average physicians per 100 000 population in Africa overall = total number of doctors (82 949)/total population (663 529) Sources: Population data from United States Census Bureau IDB Summary international Data Base (United Nations and National Statistics Offices) Number of physicians from the World Health Organization Number of FAIMER (Foundation for the Advancement of International Medical Education and Research) medical schools comes from The dependence of the United States on IMGs is encoded in various policies, most specifically Medicare's financial support for significantly more residency positions than we have domestic medical school graduates [ 10 ]. Additionally, the USA will waive the exchange visitor requirement that would otherwise return IMGs to their home countries after residency training in exchange for agreements to practice in underserved USA settings. Further, the USA will grant permanent residency status to IMGs under a variety of conditions [ 11 ]. The UK has initiated efforts to meet its own health workforce planning needs while paying attention to global equity considerations by adopting a formal "code of practice" that prohibits its National Health Service employers from recruiting health professionals from a long list of developing countries [ 12 ]. While this code has not resulted in a reduction in nurse recruitment, the number of physicians migrating to the UK has declined for a brief period (but is now back up) [ 8 , 13 ]. Recently, two prominent medical journals in the UK, the Lancet and the British Medical Journal , have editorialized on the effects of the brain drain in poor countries, recommending an international code of ethics prohibiting the recruitment of developing world health professionals by rich countries [ 14 , 15 ]. While the UK has a centralized health system well positioned to address these issues, both within its health care system and with representatives of other nations, the USA, in contrast, has a fractured health system that is less able to engage these issues. Agencies of the USA government have been reluctant, unable or unwilling to impede free-market driven physician migration. United States policies have always been quite friendly to physician migration, even taking into account toughened medical licensing examinations and tightened immigration rules over the past four or five decades. Furthermore, even though some types of immigration have been more restricted since September 11, 2001, Congress subsequently expanded the number of foreign physicians who will be granted favorable immigration status (HR 2215, passed 10/3/02 increases the number of J-1 visa waivers allocated to state health departments from 20 to 30; further, the Department of Health and Human Services took over the role formerly played by the USA Department of Agriculture in handling applications of J-1 waivers, thereby ensuring additional foreign physicians will have access to waivers.). One of the most common initial points of entry for IMG physicians into the USA medical workforce is residency training program enrollment, even if physicians have already completed postgraduate training in their home countries. The reliance of many inner-city hospitals on IMGs has thwarted calls by medical policy organizations, such as the Council on Graduate Medical Education, to reduce the number of IMGs admitted to residency programs as a means of narrowing the IMG pipeline to the USA There is little debate within the USA government or other institutions about the social justice implications of obtaining health professionals from poor countries [ 16 ]. Typically, research on the issues surrounding the role(s) of IMGs in the USA has focused on 1)whether IMGs practicing here contribute to a surplus of physician labor (which could tend to lower physician salaries and/or drive up health care costs) [ 17 - 19 ]; 2) the quality of care delivered by IMGs [ 20 ]; and 3) the contribution of IMGs to the "health safety net" in rural or underserved areas [ 21 ]. The ethics of health professional migration from poor countries to rich ones is complicated by the competition of legitimate interests – each country's need for an adequate health workforce as opposed to each individual's human right to travel. When health professionals travel to receive training and then return to apply their skills, there are advantages to the home country. Additionally, emigrants of all social classes from poor countries typically send funds home to relatives, although sub-Saharan African remittances, at less than USD 5 billion, comprise the lowest dollar amounts of any other poor world region [ 8 ]. Further, it must be noted that individuals having benefit of public funds for their medical training are sending their remittances home to private parties with no direct gain for the health or education systems. Immigration theory informs us that "push factors" prompt professionals to leave poor countries in favor of settling in higher income countries [ 22 ]. Negative factors in the sending countries include insufficient suitable employment, lower pay, unsatisfactory working conditions, poor infrastructure and technology, lower social status and recognition, and repressive governments. Simultaneously, "pull factors" in wealthier countries systematically attract physicians. These include training opportunities, higher living standards, better practice conditions and more sophisticated research conditions. The "world systems framework theory" stresses the more permeable barriers between and among countries created by the standardized curriculum and English language used in world medical schools, the use of common research methods and shared scientific knowledge, the easy articulation of requirements of practice across countries, and the weakened nationalism that occurs as a result of professional training [ 23 ]. Other theories characterize migration as a decision of family units, rather than individuals, emphasizing the insurance nature of establishing what are, in effect, "branch offices" in multiple locations [ 24 ]. Given the enduring migration from poor countries to rich ones, only likely to increase with the international liberalization of trade in health services [ 25 ], concerns for global health require the maintenance of an adequate health workforce in poor countries. Methods To describe the numbers and types of physicians practicing in the USA who earned medical degrees in Africa, we performed a cross-sectional study using the 2002 American Medical Association Physician Masterfile [ 26 ]. This data set contains detailed information on all 771 491 active physicians who were licensed to practice medicine in the year 2002 (excluding those physicians employed by federal entities such as the Veterans Administration, federal prisons or the military). We reviewed these data for all physicians in the USA who received their training in sub-Saharan Africa (those 47 countries south of the Saharan desert on the African continent). These data included year of birth, gender, year of medical school graduation, name of medical school, current practice location, specialty of practice, and practice activity (office-based, hospital-based, in residency, or conducting teaching or research). Birth country information is missing for 68% of those who graduated from a sub-Saharan African medical school, so we did not analyze birth country data. To detect changes in migrant waves over time, we analyzed the data by cohorts, categorizing physicians who had graduated from medical school during four periods: before 1970, during the 1970s, during the 1980s, and 1990 and beyond. We linked geographic data about practice locations to a four-category, rural-to-urban status and taxonomy, a condensed version of the Rural-Urban Commuting Area (RUCA) codes, to determine whether these physicians are practicing in rural or urban areas. RUCAs are a census tract-based classification scheme, that have also been adapted for zip codes, combining USA Census population data with work commuting information to characterize the types of rural and urban status [ 27 ]. Research colleagues in Canada and the UK provided some data on sub-Saharan African physicians in their countries, as well. Results A total of 179 978 (23.3%) of the 771 491 active non-federal physicians in the USA in the year 2002 received their medical qualification in another country. The largest portion of these, or 115 835 physicians, originate from low and lower-middle income nations, as defined by the World Bank. Indeed, the most frequent countries of origin of IMGs in the USA include India (36 634), the Philippines (17 755), Mexico (10 404), and Pakistan (8563). Canadian physicians conventionally are not included in the IMG count because the body that accredits USA medical schools (the Liaison Committee on Medical Education (LCME)) offers reciprocal accreditation to Canadian medical schools (accredited by the Committee on Accreditation of Canadian Medical Schools). Canadians are, however, still subject to relevant immigration requirements. Sub-Saharan African medical schools in 22 countries have trained approximately 5334 physicians currently practicing in the USA. Only nine nations, however, have lost more than 40 physicians each (see Table 2 ). Some 86% are from three countries (Nigeria, South Africa and Ghana). Nigeria, with more than twice the population of any other country in the region and 16 medical schools, has lost 2158 physicians who are now practicing in the USA; South Africa, with eight medical schools, has lost 1943 physicians; and Ghana, with three medical schools, has lost 478 physicians to the USA. By region, West Africa lost 2697 physicians and Southern Africa 1943. It is also suspected there are many more physicians from these countries working in the USA, although they are not licensed as physicians. Table 2 Country of medical school of sub-Saharan African international medical graduates (IMGs) in the United States and Canada Country of training Number of African-trained IMGs in USA 1 Number of African-trained IMGs in Canada 2 Number of physicians remaining in home country 3 % of total African-trained now in USA or Canada 4 Nigeria 2158 123 22 894 9 South Africa 1943 1845 23 844 14 Ghana 478 37 1210 30 Ethiopia 257 9 1564 15 Uganda 133 42 722 20 Kenya 93 19 4001 3 Zimbabwe 75 26 1694 6 Zambia 67 7 676 10 Liberia 47 8 72 43 Other 12 countries 83 35 12 912 1 Total/Average 5334 2151 69 589 10 1. American Medical Association: Physicians' professional record (AMA-PPD). 2002 2. Buske, Lynda. Associate director of research, Canadian Medical Association. Personal communication. February 3, 2003. 3. Number of physicians from the World Health Organization. Available at: Calculation: [(Col. 1 + Col. 2)/ (Col. 1 + Col. 2 + Col. 3)]100 = percent Other 12 countries with at least one graduate in the United States. An analysis by school indicates only ten medical schools produced 79.4% of the sub-continent's graduates who are practicing in the USA. The medical schools most frequently attended by Sub-Saharan African IMGs in the USA include the University of the Witwatersrand (South Africa, 1053 physicians), the University of Cape Town (South Africa, 655), the University of Ibadan (Nigeria, 643), the University of Lagos (Nigeria, 429), the University of Nigeria (Nigeria, 394), the University of Ghana (Ghana, 389), Addis Ababa University (Ethiopia, 200), the University of Benin (Nigeria, 183), the University of Ife (Nigeria, 156), and the University of Pretoria (South Africa, 132), for a total of 4234 physicians. An analysis of the numbers of sub-Saharan African IMGs coming to the USA in each of the last decades illustrates that it takes some time between graduation and emigration. The number of recent graduates currently in a USA residency program is higher than those in previous decades because of the obvious correlation between age and career stage. Among sub-Saharan physicians in the USA, 78.3% are male. The picture is changing over time, however. Of the cohort who were trained in 1969 or earlier, 90% were male, but now only 66.3% of those who graduated from medical school in 1990 or later are male (see Table 3 ). Table 3 Characteristics of sub-Saharan African international medical graduates in the United States by graduation year cohort 1969 or earlier 1970–1979 1980–1989 1990–2000 OVERALL Number 720 1167 2268 1179 5334 Currently in residency (%) 0 2.3 17.6 59.7 21.2 Gender (% male) 90.0 86.5 76.7 66.3 78.3 Generalists (%) 19.0 28.3 47.8 57.9 41.9 current practice location: Urban (%) 95.3 94.1 93.7 95.7 94.4 Large rural (%) 2.4 3.3 3.6 2.6 3.1 Small rural (%) 1.7 2.1 1.8 1.1 1.7 Isolated rural (%) 0.7 0.6 0.9 0.6 0.7 Note: includes residents. Source of data: American Medical Association: Physicians' professional record (AMA-PPD) 2002. The average age of sub-Saharan African physicians in the USA is 43 years, compared to 46 years for all USA physicians. Forty two percent of sub-Saharan African physicians in the USA are under 40 years, and another 32% are between 40 and 50. Among the large contributing countries, Nigerian physicians are the youngest cohort (63% are under 40), and South Africans are the oldest (only 20% are under 40). A higher proportion of sub-Saharan physicians were in residency training programs (21.2%) than were USA physicians (14.1%), because many emigrate specifically for that reason. While 41.9% are in generalist specialty areas, compared to 34.7% of USA-trained physicians, the number has been rising with each new cohort. Table 3 illustrates that 57.9% of those trained in the 1990s selected a generalist practice specialty, compared to 28.3% of those trained in the 1970s. This apparently rising interest in generalist practice may be an artifact, however, as a prerequisite to internal medicine specialization is training in general internal medicine. While 31.6% of all sub-Saharan African physicians in the USA are identified as family practitioners or general internists, it may be that this ratio of generalists will increase, as 45.4% of those in residency programs are in those two specialties. The next largest specialty groups are pediatrics (9.7%), psychiatry (5.5%), anesthesiology (5.4%), obstetrics and gynecology (3.3%) and general surgery (3.0%). Urban areas attracted 93% of sub-Saharan African physicians (compared to 90.9% of other IMGs), even after excluding residents, who are typically based in urban teaching hospitals. Graduates of USA medical schools distribute themselves similarly, with 87% of USA-trained physicians in urban areas, even though a smaller 81% of the population lives in urban areas [ 27 ]. The states attracting the largest numbers of sub-Saharan Africa physicians include New York, California, Texas, Maryland, Illinois, Georgia, Pennsylvania, and New Jersey (see Figure 1 ). These are the same states that draw the largest portion of immigrant physicians generally. Figure 1 Origin and distribution of African-trained physicians in the 11 US states with the most such physicians The 1943 physicians trained in South African medical schools are somewhat different from their fellow African trainees from the subcontinent. They are older, more of them are male, they are typically white (94%) and they are more often in a subspecialty practice. This may reflect a particular wave of physicians seeking subspecialty training and practice opportunities in the USA during the political turmoil South Africa experienced in the 1970s and 1980s. It is unlikely they are seeking training abroad that is unavailable in their home country, as the medical training opportunities in South Africa are quite comprehensive. Discussion The 5000-plus physicians trained in sub-Saharan Africa who have migrated to the USA comprise only a small proportion of the total number of IMGs practicing in the USA. However, their relatively small numeric addition to the USA medical workforce contrasts markedly with the impact of their migration on the medical workforce in sub-Saharan Africa. Moreover, in absolute terms, the USA has drawn more of the medical workforce of Africa than either Canada, with 2151 African graduates (Lynda Buske, Canadian Medical Association, personal communication, 2/3/03), or the UK, with 3451 (Bonnie Sibbald, University of Manchester, personal communication, 1/24/03), largely because of the relative size of the health care system in the USA. Including the USA, the UK, and Canada, then, 10 936 physicians trained in sub-Saharan Africa are practicing in the three countries, a number that represents 12% of all African physicians. We expect there are many more African physicians who are in the UK, as well, as our figures from there include only those who arrived post-1992. While some of the physicians in residency training will return home, there are unknown others who could be practicing medicine at home but were not able to get licenses abroad and therefore are engaged in other occupations. Almost all the medical schools sending graduates to the USA provide their instruction in English. Our study provides some measures of sub-Saharan Africa physician migration to the USA. A next step is to collect data from other developed countries and begin to create a physician migration data set from multiple countries. Migration from country to country within the subcontinent is also worthy of further examination, and it would be useful to identify the relationship between country of birth and country of training. There are likely to be some medical schools that draw students from several African countries. It should be noted, as well, that there are physicians practicing in Africa who did not train there, most prominently Cubans. Some sub-Saharan countries lose a larger proportion of their physicians to the USA than others. For example, while Ghana has a reported 1210 practicing physicians in its country, 478 graduates of Ghanaian medical schools are practicing in the USA. Without even considering those who have migrated to other countries, these 478 Ghanaian graduates in the USA represent 30% of Ghana's potential medical workforce (see Table 2 ). If none of those had come to the USA, the physician-to-population ratio in Ghana would rise from 6.2 to 8.7 per 100 000, or a 40% increase. By comparison, South Africa has lost 14% of its potential workforce to the USA and Canada. The migration of physicians from sub-Saharan Africa represents a lost investment of significant training costs, since graduates of medical schools in Africa are likely to have contributed financially to only a small portion of the costs of their medical education [ 28 ]. Medical education is estimated to cost Ghana about USD 9 million per year and Nigeria USD 20 million (Hagopian & Ofosu, et al. The flight of physicians from West Africa: views of African physicians and implications for policy , unpublished 2003). The United Nations Commission for Trade and Development has estimated that each professional leaving Africa costs the continent USD 184 000, or USD 4 billion a year – one third of official development funds to Africa [ 29 ]. The loss of trained health personnel also contributes to a general decline in average incomes, as physicians generate skilled health system jobs beyond their own. Lost tax revenues from absent physicians represent significant losses as well. Ostensibly, the USA welcomes IMGs for two purposes. First, as a form of foreign aid, it provides specialty training that physicians can take back to their home countries for the benefit of residents of those nations. Second, IMGs fill positions in specialties and locations that are less attractive to their USA counterparts, and may help to correct physician maldistribution in some rural or underserved areas of the USA. (There are several federal agencies, along with state health departments who "sponsor" physicians who have completed their residency training in the USA on J-1 exchange – or "student"- visas. These sponsorships allow foreign national physicians to gain approval from the State Department and the USA Citizenship and Immigration Services to waive J-1 visa requirements that would otherwise require them to return to home countries for at least two years. In exchange for this waiver, physicians find employment with a health agency or private physician in a health professional shortage area.) Longitudinal tracking of physicians entering the USA has indicated, however, that few IMGs ever leave the USA after arriving for residency training [ 30 ], and there is conflicting evidence about whether IMGs are more likely to practice in safety net practices for low-income and underinsured people [ 21 , 31 , 32 ]. Attracting physicians to rural practice in most countries is difficult, and is accomplished only through a careful set of policies designed to provide incentives for rural service. In most poor as well as rich countries, physicians are concentrated around urban hospitals that offer tertiary care, even though more rational service delivery systems might focus on a geographically decentralized system of primary and preventive care. Poor countries that offer medical training to produce too many physicians with highly technical skills, some of whom who cannot find satisfying jobs, may further contribute to physician migration [ 33 ]. India and the Philippines, for example, clearly over-produce physicians who are intended for an international market. Our findings show African physicians are unlikely to select small or remote rural practice opportunities in either their home countries or in the USA, but the preponderance of African physicians in American inner-city underserved areas may to some extent be helpful to USA needs by boosting the number of minority physicians in the urban health workforce. The growing number of African immigrant female physicians follows the trend for increasing numbers of female physicians trained in the USA, and probably has similar implications. Some researchers have found, for instance, that female physicians are less likely to practice in rural areas [ 34 ]. While the sub-Saharan Africa region as a whole loses many of its physicians, it is apparent that a small handful of medical schools is the sources of the majority of this migration. Ten medical schools in four countries – South Africa, Nigeria, Ghana and Ethiopia – produce 79.4% of the émigré physicians to the USA, out of a total of 87 medical schools in the region. This suggests policy approaches to reducing the "brain drain" from Africa could be targeted at only these few countries or medical schools, a less daunting task than addressing the problem in 47 different countries. Medical migration results from the complex interaction of myriad social, legal and economic forces. Single country policies are unlikely to alter the flows significantly. Even if the USA acknowledges that that it benefits from luring medical professionals here for whose medical school training we do not pay, solutions that would be compatible with social justice principles are not clear. Furthermore, if all African doctors returned to their home countries today, they would not necessarily find satisfactory employment opportunities in cash-strapped health systems. Conclusions The 57th World Health Assembly, in 2004, adopted a resolution to urge member countries to develop strategies to mitigate the adverse effects of migration of health workers; to develop policies that could provide incentive for health workers to remain in their countries; and, among other issues, requests WHO to help countries set up information systems to monitor the movement of health resources for health, and to include human resources for health development as a top-priority program at WHO from 2006 to 2015 [ 35 ]. In an ideal world, freedom of movement is a universal right for individuals, as there is ostensibly no rational reason why anyone would have a stronger right to be in any place more than anyone else [ 36 ]. Today, however, differences in wealth between countries create flows of educated people seeking better opportunities far from home. One result is that resource-strapped African (and other poor) countries have invested significant resources in educating health professionals who will never serve the populations that were taxed (or took out high-interest loans from international lenders) to pay for their training. Importing health professionals from poor countries to provide care in rich countries is not consistent with a rational workforce policy rooted in social justice principles. In the short run, Mullan [ 37 ] and others are right to recommend that the USA expand its incentives to USA graduates to practice in rural and underserved areas through the National Health Service Corps and other programs. Grumbach [ 18 ] recommends reducing the number of excess residency training positions by limiting the Medicare subsidy. In response to a recent report by the USA Physicians for Human Rights [ 38 ], The New York Times editorialized that the "obvious long-term solution to the medical brain drain is for wealthier countries to reimburse Africa's health and educational systems for the cost of poaching their professionals, and to greatly increase the financing and technical help for Africa's health systems" [ 39 ]. This unprecedented attention to the issue of the African medical brain drain in a major USA publication, coupled with a radical call for reparations, suggests USA policy makers may be called to address this issue. The same Physicians for Human Rights report that prompted The New York Times response made a strong recommendation that the International Monetary Fund, World Bank, and other donors refrain from withholding loans or grants from countries that increase their spending on "health, education and other sectors and activities needed to promote human development, including to enhance salaries to health staff or to hire new health personnel." One of the major limitations African nations face in addressing their health workforce problems is the lack of reliable data on how many health workers have graduated from their schools, how many are working in the country and in what locations, and how many have emigrated. It is urgent that poor countries put together the information systems required to track these data, as a basis for workforce policy and investment decisions. And, finally, the fact that so few medical schools generate the vast majority of emigrants creates an opportunity to focus attention in a strategic way. These schools might be enticed to redirect their missions towards producing graduates who intend to serve their own countries. This would likely require curriculum changes, admissions policy changes, and a change in faculty culture to ensure that emigration is not promoted as a mark of prestige. Competing interests The author(s) declare that they have no competing interests. Authors' contributions AH conceived the project, designed the research and wrote the paper. MJT and KEJ actively participated in the conceptualization and re-writing of the paper. MF conducted the data management and analysis. LGH provided guidance, advice and editing assistance. All authors read and approved the final manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC544595.xml
514605	A study comparing the actions of gabapentin and pregabalin on the electrophysiological properties of cultured DRG neurones from neonatal rats	Background Gabapentin and pregabalin have wide-ranging therapeutic actions, and are structurally related to the inhibitory neurotransmitter GABA. Gabapentin, pregablin and GABA can all modulate voltage-activated Ca 2+ channels. In this study we have used whole cell patch clamp recording and fura-2 Ca 2+ imaging to characterise the actions of pregabalin on the electrophysiological properties of cultured dorsal root ganglion (DRG) neurones from neonatal rats. The aims of this study were to determine whether pregabalin and gabapentin had additive inhibitory effects on high voltage-activated Ca 2+ channels, evaluate whether the actions of pregabalin were dependent on GABA receptors and characterise the actions of pregabalin on voltage-activated potassium currents. Results Pregabalin (25 nM – 2.5 μM) inhibited 20–30% of the high voltage-activated Ca 2+ current in cultured DRG neurones. The residual Ca 2+ current recorded in the presence of pregabalin was sensitive to the L-type Ca 2+ channel modulator, Bay K8644. Saturating concentrations of gabapentin failed to have additive effects when applied with pregabalin, indicating that these two compounds act on the same type(s) of voltage-activated Ca 2+ channels but the majority of Ca 2+ current was resistant to both drugs. The continual application of GABA, the GABA B receptor antagonist CGP52432, or intracellular photorelease of GTP-γ-S had no effect on pregabalin-induced inhibition of Ca 2+ currents. Although clear inhibition of Ca 2+ influx was produced by pregabalin in a population of small neurones, a significant population of larger neurones showed enhanced Ca 2+ influx in response to pregabalin. The enhanced Ca 2+ influx evoked by pregabalin was mimicked by partial block of K + conductances with tetraethylammonium. Pregabalin produced biphasic effects on voltage-activated K + currents, the inhibitory effect of pregabalin was prevented with apamin. The delayed enhancement of K + currents was attenuated by pertussis toxin and by intracellular application of a (Rp)-analogue of cAMP. Conclusions Pregabalin reduces excitatory properties of cultured DRG neurones by modulating voltage-activated Ca 2+ and K + channels. The pharmacological activity of pregabalin is similar but not identical to that of gabapentin. The actions of pregabalin may involve both extracellular and intracellular drug target sites and modulation of a variety of neuronal conductances, by direct interactions, and through intracellular signalling involving protein kinase A.	Background Gabapentin (Neurotonin ® ) and pregabalin (S(+)-3-isobutyl GABA) were both originally designed as GABA mimetics (Figure 1A ), with the intention that they would be able to cross the blood-brain barrier and interact with GABAergic systems and enhance GABA mediated inhibition. Although gabapentin appears to have diverse therapeutic utility in the treatment of pain disorders [ 1 ], psychiatric illnesses [ 2 ] and epilepsy [ 3 ], there is controversy regarding its molecular mechanisms of action. Whether the actions of gabapentin and pregabalin are mediated through GABAergic mechanisms or GABA receptors remains particularly contentious [ 4 - 6 ]. However, high affinity binding sites for gabapentin and pregabalin on distinct α 2 δ subunits of voltage-activated calcium channels have been identified and characterised [ 7 ]. For this reason voltage-activated Ca 2+ channels remain primary candidate sites of action for these novel anticonvulsant and antihyperalgesic drugs. Functional data from studies on gabapentin and pregabalin also support this contention. Specifically, both gabapentin and pregabalin inhibited hyperalgesia [ 8 , 9 ], attenuated evoked Ca 2+ influx into brain slices and reduced evoked transmitter release [ 10 ]. Additionally, gabapentin and pregabalin inhibited multiple firing of action potentials evoked by 300 ms depolarising current commands in cultured sensory DRG neurones [ 5 ]. Figure 1 Pregabalin inhibits Ca 2+ currents. A) Structure of GABA (γ-aminobutyric acid), gabapentin (1-(aminoethyl)cyclohexane acetic acid) and pregabalin (S(+)-3-isobutyl GABA). B & C) Traces of high voltage-activated Ca 2+ currents evoked from a holding potential of -90 mV by a depolarising step command to 0 mV showing inhibition by pregabalin (2.5 μM). B) Shows that pressure ejection of choline chloride extracellular solution does not induce inhibition of the Ca 2+ current but in the same neurone pregabalin does produce a response. Traces show a control Ca 2+ current (Ctrl), the current unaffected by application of choline chloride recording solution (ChCl), inhibition of current by 3 minutes application of pregabalin (PGB) and the current at 5 minutes recovery (Rec). C) Traces show a control Ca 2+ current (Ctrl), the inhibition of current after 3 minutes application of pregabalin (PGB) and full recovery of the current 5 minutes after removal of the drug pipette (Rec). D) Graph showing the distribution of inhibitory responses produced by 0.025 – 2.5 μM pregabalin. Each symbol represents a result from a different experiment. Our previous studies have focused on the inhibitory effects of gabapentin (0.25–25 μM) on whole cell voltage-activated Ca 2+ currents and K + stimulated Ca 2+ entry measured with fura-2. The cellular model systems used were primary cultures of dorsal root ganglion (DRG) neurones from 1–4 day old rats and differentiated F-11 cells (embryonic rat DRG × neuroblastoma hybrid cell line) [ 11 , 12 ]. In this present study the effects of pregabalin and gabapentin on the electrophysiological properties of cultured neonatal rat DRG neurones were measured with particular reference to Ca 2+ entry through high voltage-activated channels and enhancement of K + conductances. We had two specific aims for this project. The first was to determine whether gabapentin and pregabalin have the same mechanisms of action, by examining whether saturating concentrations of gabapentin and pregabalin act in an additive manner to attenuate Ca 2+ influx. The second aim was to further evaluate GABA receptors as target sites for these drugs in sensory neurones. This latter element to the study was conducted because it has been proposed that specific GABA B receptors with a gb1a-gb2 heterodimer composition are the sites of agonist activity of gabapentin. These specific receptors can be coupled to an inwardly rectifying K + channel and / or voltage-activated Ca 2+ channels to dampen neuronal electrical excitability [ 13 - 15 ]. However, in certain situations gabapentin and the GABA B receptor agonist, baclofen, have different actions. Weaver mutant mice (wv/wv) are insensitive to both gabapentin and baclofen, however in control littermates, Weaver control mice (+/+, wv/+), only baclofen evoked a K + current [ 16 ]. Results and discussion Actions of pregabalin on voltage-activated calcium currents The firing of multiple action potentials in response to a sustained depolarising current command is a property of a sub-population (under 20 %) of cultured DRG neurones. Application for 3 minutes, of either pregabalin (PGB; 2.5 μM) or gabapentin (GBP; 2.5 μM) reduced the frequency of action potential spikes evoked by 300 ms depolarising current step commands applied every 30 s [ 5 ]. Gabapentin attenuated repetitive action potential firing as measured by a reduction in the mean number of evoked action potentials during 300 ms depolarisations from 7 to 2 (n = 3, p < 0.01 ). Pregabalin (2.5 μM) reversibly reduced the number of action potentials during 300 ms depolarisations from 8 to 3 (n = 5, p < 0.01 ). However, pregabalin and gabapentin did not significantly alter the properties (amplitude, duration & threshold) of single action potentials evoked by 5 ms depolarising current commands. Consistent with our previous work [ 11 , 12 ], whole cell voltage-activated Ca 2+ currents (I Ca ) recorded from DRG neurones were reversibly attenuated (22 ± 11%; n = 7) by 2.5 μM gabapentin. Similarly, 3 minutes application of pregabalin (25 nM – 2.5 μM; Figure 1B & 1C ; table 1 ) inhibited the mean Ca 2+ current amplitude, measured at the peak of the inward current, and at the end of a 100 ms voltage step command to 0 mV. The inhibitory actions of pregabalin were not accompanied by any shift in the voltage-dependence of activation for I Ca or by any change in holding or leak currents. At least partial recovery of I Ca was observed 5 minutes after removal of the pregabalin-containing perfusion pipette (Figure 1B & 1C ). A very low concentration of pregabalin (0.25 nM) failed to produce significant inhibition of I Ca . However, no clear dose-dependent relationship was established for the inhibitory actions of pregabalin, and a major part of the voltage-activated Ca 2+ current was insensitive to the drug. It was also apparent that there was considerable variability in the sensitivity of the DRG neurones to any given dose of pregabalin. Some neurones did not respond to 2.5 μM pregabalin while in a few neurones this same concentration produced 60 % inhibition of I Ca (Figure 1D ). Table 1 Inhibitory actions of pregabalin on voltage-activated calcium currents. Peak control current amplitude (nA) Pregabalin Concentration (μM) Peak current in the presence of pregabalin (nA) Percentage inhibition and n value -1.52 ± 0.13 0.025 -0.98 ± 0.12 ** 31 ± 4 % (n = 8) -1.2 ± 0.14 0.25 -0.93 ± 0.18 * 22 ± 9 % (n = 6) -1.19 ± 0.11 2.5 -0.94 ± 0.09* 21 ± 3 % (n = 26) P < 0.001; P < 0.01 and P < 0.05. Each value is given ± S.E.M. Data shows inhibition but a lack of dose-dependent actions on I Ca of a wide range of pregabalin concentrations. Although varied responses were observed it was clear that a major component of I Ca was resistant to inhibition by pregabalin. Distinct voltage-activated Ca 2+ channels, defined by their alpha 1 subunit, have been suggested to be selective target sites for gabapentin and related drugs like pregabalin. With this in mind, the 1,4-dihydropyridine L-type Ca 2+ channel agonist Bay K8644 was used to determine whether pregabalin was inhibiting L-type channels. Previously, we found that in the continual presence of gabapentin, Bay K8644 enhanced I Ca [ 12 ]. In this present study similar results were obtained. After inhibiting part of the current with pregabalin (2.5 μM), Bay K8644 (1 μM) was applied with pregabalin and the enhanced I Ca was measured at its peak and at the end of a 100 ms voltage step command to 0 mV (Figure 2 ). The percentage increases in current seen with Bay K8644 in the presence of either gabapentin or pregabalin were slightly less than those seen under control conditions in cultured DRG neurones [ 17 ]. This is consistent with some L-type Ca 2+ current modulation by pregabalin and gabapentin in DRG neurones. Taken together these data suggest that neither gabapentin nor pregabalin are selectively inhibiting L-type I Ca in DRG neurones. This contrasts with the effects of gabapentin on cortical pyramidal neurones where inhibition of L-type Ca 2+ channels appears to be the predominant mechanism of action [ 18 ]. Figure 2 Bay K8644 enhanced pregabalin-insensitive current suggesting that L-type Ca 2+ channels are still available for modulation by the 1,4-dihydropyridine agonist. A) Bar chart shows data for mean calcium current amplitudes measured at the peak of the inward current (open bars) and at the end of a 100 ms voltage step command to 0 mV (solid bars, n = 6). Data under control condition (Con) in the presence of 2.5 μM pregabalin (PGB) and in the presence of both pregabalin (2.5 μM) and Bay K8644 (1 μM) are shown. B) The inset traces show voltage and Ca 2+ current records under control conditions, inhibition in the presence of 2.5 μM pregabalin (PGB) and enhancement of the Ca 2+ current during continued application of pregabalin with Bay K8644 present (Bay K & PGB). C) Shows the current inhibited by pregabalin, (obtained by subtracting the net current recorded in the presence of pregabalin from the net control current). D) Shows the additional current produced by Bay K8644, (obtained by subtracting the net current recorded in the presence of pregabalin from the net current recorded in the presence of both pregabalin and Bay K8644). Actions of pregabalin on calcium influx through voltage-activated channels, measured using fura-2 imaging Given the variable and rather modest but reversible inhibitory actions of pregabalin on I Ca we tested the actions of pregabalin on K + -evoked Ca 2+ influx using fura-2 imaging. An extracellular solution containing 30 mM K + was used to depolarise the DRG neurones and activate three consistent Ca 2+ transients [ 11 , 12 ]. Pregabalin (2.5 μM) was applied during the second K + -evoked depolarisation and it produced a mixture of reversible effects on the Ca 2+ transients. In 4 of 24 neurones the Ca 2+ flux was decreased by 54 ± 13 % ( p <0.05) but in 20 neurones from the same cultures pregabalin evoked a mean increase in Ca 2+ flux to 185 ± 20 % ( p <0.05) of the control. Raising the pregabalin concentration to 25 μM increased the proportion of inhibitory responses (n = 21 out of 49 neurones) but enhancement of Ca 2+ transients was still observed in the remaining 28 neurones (Figure 3 ). However, 250 μM pregabalin caused an increase in K + -evoked Ca 2+ influx in all neurones studied (n = 31). Although in some cells good recovery from pregabalin actions was observed (figure 3B ) the inhibitory effect was often not associated with recovery. This may reflect long lasting effects of pregabalin. Run-down in some cells can not be completely ruled out but under control conditions the mean level of run-down of three K + -evoked Ca 2+ transients was only 8.4 ± 1.5 % (n = 39). Similarly, the mean control level of increase in K + -evoked Ca 2+ transients was only 7.2 ± 1.3 % (n = 35). Figure 3 Pregabalin produced mixed actions on Ca 2+ influx evoked by 30 mM K + . A) Bar chart showing inhibition of Ca 2+ influx by pregabalin (25 μM). Data for the total Ca 2+ fluxes was normalised with respect to the first control response to K + (Ctrl). The second and third responses were obtained in the presence of pregabalin (PGB) and after washing away the pregabalin (Rec). B) Record of K + -evoked Ca 2+ transients showing partially reversible inhibition produced by 25 μM pregabalin (PGB). The period of stimulation is shown with open bars and PGB application with the filled bar. C) Bar chart showing enhancement of Ca 2+ influx by pregabalin (25 μM). Data for the total Ca 2+ fluxes were normalised with respect to the first control response to K + (Ctrl). The second and third responses were obtained in the presence of pregabalin (PGB) and after washing away the pregabalin (Rec). D) Record of K + -evoked Ca 2+ transients showing reversible enhancement produced by 25 μM pregabalin (PGB). To investigate the different responses to pregabalin, the sizes of cell somas were measured and compared. Although there is overlap, figure 4 shows that enhancement of K + -evoked Ca 2+ transients by pregabalin was mainly seen in neurones with larger cell somas and that in smaller neurones pregabalin produced inhibitory effects. Figure 4 Different responses to pregabalin were observed in different populations of cultured DRG neurones. Bar chart showing the distribution of neurones with different cell soma areas and the response to pregabalin. The distributions for intermediate and larger neurones where pregabalin (25 μM) increased K + -evoked Ca 2+ influx are shown in black bars. The distributions for small and some intermediate neurones where pregabalin (25 μM) attenuated K + -evoked Ca 2+ influx are shown in open bars. Ca 2+ -dependent conductances of DRG neurones are sensitive to ryanodine and Ca 2+ -induced Ca 2+ release has been reported in these neurones. Modulation of either Ca 2+ -induced Ca 2+ release and / or Ca 2+ homeostatic mechanisms might provide a mechanism by which pregabalin enhanced Ca 2+ transients in neurones with intermediate and large sized cell somas. This was investigated in two ways. Firstly, the actions of pregabalin on caffeine-evoked Ca 2+ transients were evaluated in nominally Ca 2+ -free extracellular conditions (NaCl-based solution with no added CaCl 2 ). Single caffeine (1 mM) responses were obtained from DRG neurones in either the absence or presence of 25 μM pregabalin. No differences in either amplitudes or durations of Ca 2+ transients were seen when 8 control caffeine responses were compared with 4 caffeine responses obtained in the presence of pregabalin (Figure 5A,5B ). The second approach was to explore a possible role of Na + / Ca 2+ exchange by bathing cells in choline chloride-based medium containing only 1 mM Na + . Under these conditions the contribution of the Na + / Ca 2+ exchanger to handling of intracellular Ca 2+ loads will be minimal. In choline chloride-based medium pregabalin produced both enhancement and inhibition of total Ca 2+ flux in 7 and 3 neurones respectively. Enhancement in Ca 2+ flux by pregabalin under these experimental conditions, suggest that inhibition of Na + / Ca 2+ exchange is not the main mechanism by which pregabalin enhances K + -evoked Ca 2+ flux (Figure 5C ). However, detailed analysis was made difficult by poor recovery of even the first Ca 2+ transient evoked in low extracellular Na + which does suggest that Na + / Ca 2+ exchange is an important homeostatic mechanism in DRG neurones. In conclusion these results indicate that modulation of Ca 2+ -induced Ca 2+ release and Ca 2+ homeostatic mechanisms do not account for pregabalin-induced enhancement of K + -evoked Ca 2+ flux in a population of DRG neurones. Figure 5 Pregabalin does not appear to modulate caffeine-evoked Ca 2+ release or Ca 2+ homeostatic mechanisms. A & B) Show example records of Ca 2+ transient evoked by caffeine (1 mM) applied to DRG neurones bathed in nominally Ca 2+ -free medium and measured using fura-2. Under these conditions the Ca 2+ transients are only due to mobilisation of Ca 2+ from intracellular stores. A) Illustrates a control caffeine response and B) shows a similar response to caffeine recorded in the presence of 25 μM pregabalin (PGB). C) Example trace showing an increase in K + -evoked Ca 2+ flux by 25 μM pregabalin (PGB) in a DRG neurone bathed with choline chloride-based extracellular medium. The periods of stimulation with 30 mM KCl are shown with open bars and pregabalin application is shown with a filled bar. D) Example trace showing an increase in K + -evoked Ca 2+ flux induced by 5 mM TEA in a DRG neurone bathed with standard NaCl-based extracellular medium. The periods of stimulation with 30 mM KCl are shown with open bars and TEA application is shown with a filled bar. Inhibitory modulation of potassium conductances could result in increased K + -evoked Ca 2+ transients. To test this alternative mechanism, a relatively low concentration, 5 mM, of tetraethylammonium (TEA) was applied with NaCl-based extracellular medium and DRG neurones were stimulated with 30 mM KCl. Attenuation of potassium conductances by TEA markedly increased the K + -evoked Ca 2+ flux in all eight DRG neurones studied, mimicking in part the action of pregabalin (Figure 5D ). Do pregabalin and gabapentin have additive effects on cultured dorsal root ganglion neurones? Pregabalin and gabapentin have related chemical structures and appear to have similar but usually modest inhibitory effects on Ca 2+ currents. Both pregabalin and gabapentin at a concentration of 2.5 μM produced a maximum level of current inhibition. Simultaneous application of pregabalin (2.5 μM) and gabapentin (2.5 μM) produced modest but significant inhibition of I Ca at the peak of the current and at the end of the stimulus (Figure 6A ). However, the percentage inhibition of I Ca produced was not significantly different for 2.5 μM gabapentin alone (22 ± 11%; n = 7), 2.5 μM pregabalin alone (21 ± 3%; n = 26) and 2.5 μM gabapentin and 2.5 μM pregabalin applied together (19 ± 2%; n = 9). Figure 6 Pregabalin and gabapentin do not have additive actions on cultured DRG neurones. A) Bar chart showing the mean Ca 2+ current amplitude recorded at the peak of the inward current (Peak) and end of a 100 ms voltage step command (End). Data is shown for measurements made under control conditions (Control), after 3 to 5 minutes application of both pregabalin (2.5 μM) and gabapentin (2.5 μM) (PGB + GBP) and after 5 minutes recovery (Recovery). B) Bar chart showing normalised data from Ca 2+ imaging experiments in which simultaneous application of both pregabalin (25 μM) and gabapentin (25 μM) (PGB & GBP) reversibly attenuated the total Ca 2+ flux. C) Record of K + -evoked Ca 2+ transients, showing the reversible inhibition produced by 25 μM pregabalin and 25 μM gabapentin (PGB + GBP, filled bar). In spite of the added complexity of mixed responses with pregabalin, experiments were carried out using fura-2 to measure K + evoked Ca 2+ flux and the effects of both ligands. Our previous experiments with gabapentin (25 μM) had only identified inhibitory effects using this experimental approach [ 11 , 12 ]. Simultaneous application of pregabalin and gabapentin (both at 25 μM) reversibly reduced the K + evoked Ca 2+ influx (Figure 6B & 6C ). Pregabalin (25 μM) and gabapentin (25 μM) together reduced the total Ca 2+ flux to 75 ± 5 % (n = 14) of the control response to 30 mM K + . Consistent with the electrophysiological experiments, this level of inhibition was similar to the inhibitory effects of pregabalin and gabapentin applied separately. The electrophysiological and fura-2 Ca 2+ imaging data show that no additive inhibitory effects were found during simultaneous application of saturating concentrations of gabapentin and pregabalin. The data therefore support the contention that both drugs have a closely related or a common site of inhibitory action. The data also indicate that in the modulation of Ca 2+ channels these drugs do not act in an additive manner even though a substantial proportion of current is resistant to both drugs. It should be emphasized however that pregabalin can reversibly enhance K + -evoked Ca 2+ transients, an effect not seen with gabapentin. Therefore there may be other additional actions of pregabalin that can be identified in Ca 2+ imaging experiments when all membrane conductances are intact. Does GABA receptor modulation alter responses to pregabalin in cultured dorsal root ganglion neurones? In this section of the study two strategies were used to evaluate the possible roles of GABA receptors in pregabalin responses in cultured DRG neurones. Firstly, pregabalin actions were studied in the presence of a saturating concentration of GABA; this initially activated all types of GABA receptor and then caused desensitization of these receptors. Secondly, the effect of a potent and selective GABA B receptor antagonist, CGP52432 [ 19 ] on pregabalin responses was evaluated. GABA (100 μM) evoked inward currents in a sub-population of cultured DRG neurones and induced a transient rise in intracellular Ca 2+ in 18 of 41 DRG neurones studied (Figure 7A & 7B ). The inward currents were due to the activation of GABA A receptor Cl - channels; with the equilibrium potential for Cl - under our recording conditions being close to 0 mV the chloride conductance results in an inward current. Although in vivo the equilibrium potential for Cl - in DRG neurones is variable, it is predicted to be between -40 mV and -20 mV because of Cl - loading into the intracellular environment (for review see [ 20 ]). Therefore GABA A receptor Cl - channel activation can produce a depolarisation of the resting membrane potential. Our data indicated that the GABA-evoked depolarisation was sufficient to result in voltage-activated Ca 2+ channel activity and produce a transient rise in intracellular Ca 2+ . In both neurones that responded to GABA and those that did not, subsequent application of pregabalin, produced either enhancement or inhibition of K + -evoked Ca 2+ influx (Figure 7A,7B,7C & 7D ; Table 2 ). These responses to pregabalin in the presence of GABA were no different in character to the pregabalin responses obtained in the absence of GABA. In neurones that showed a Ca 2+ transient in response to a GABA-evoked depolarisation, there was an apparent increase (to 83%) in the proportion of neurones that were inhibited by pregabalin. It is not clear why this is but it may reflect the sensitivity of Ca 2+ transients evoked in DRG neurones, which can only be evoked consistently in most DRG neurones by three depolarising stimuli. Thus the GABA-evoked responses may influence Ca 2+ homeostatic mechanisms and subsequent stimulated Ca 2+ entry. However, the critical observation from these experiments is that GABA receptor desensitisation does not prevent either Ca 2+ transient enhancement or inhibitory actions of pregabalin in neurones that were sensitive or insensitive to GABA. Figure 7 Activation and subsequent desensitisation of GABA receptors fails to prevent the modulation of voltage-activated Ca 2+ channels in cultured DRG neurones by pregabalin. A) Bar chart showing data obtained from neurones that responded to GABA (100 μM). All data are normalised with respect to the first Ca 2+ transient evoked by 30 mM KCl. Open bars show the relative responses to 100 μM GABA in cells (n = 3) where 25 μM pregabalin enhanced the Ca 2+ flux. Filled bars show responses in cells (n = 15) where 25 μM pregabalin inhibited the Ca 2+ flux. Data are shown for GABA responses (GABA), K + -evoked Ca 2+ transients under control conditions with GABA present (Ctrl), K + evoked Ca 2+ transients in the presence of GABA and 25 μM pregabalin (PGB) and on recovery in the continued presence of GABA (Rec). B) An example trace, showing a response to 100 μM GABA and Ca 2+ transients evoked by K + in the presence and absence 25 μM pregabalin. The long open bar shows the period of GABA application, the short open bars show K + stimulation and the filled bar the period of pregabalin application. C) Bar chart showing data obtained from neurones that did not responded to GABA (100 μM) but were continually bathed with GABA for the duration of the experiment. All data are normalised with respect to the first Ca 2+ transient evoked by 30 mM KCl. Data are shown for K + -evoked Ca 2+ transients under control conditions with GABA present (Ctrl), increased (n = 10; open bars) and decreased (n = 13; filled bars) K + -evoked Ca 2+ transients in the presence of GABA and 25 μM pregabalin (PGB) and on recovery (Rec). D) An example trace, showing no response to 100 μM GABA but responses to K + in the presence and absence of 25 μM pregabalin. The long open bar shows the period of GABA application, the short open bars show K + stimulation and the filled bar the period of pregabalin application. Table 2 Actions of pregabalin (25 μM) on K + -evoked Ca 2+ influx in the continual presence or absence of 100 μM GABA. GABA sensitivity Mean percentage of control Ca 2+ influx Number of neurones Proportion of Neurones Responders 110 ± 2 % 3 of 18 17 % Responders 77 ± 2 % 15 of 18 83 % Non-responders 121 ± 4 % 10 of 23 43 % Non-responders 81 ± 3 % 13 of 23 57 % Control 169 ± 24 % 28 of 49 57 % Control 85 ± 2 % 21 of 49 43 % Responders were those neurones in which GABA evoked a transient rise in intracellular Ca 2+ . Non-responder showed no change in fura-2 fluorescence ratio in response to GABA. Controls where neurones that pregabalin was applied to without GABA being present. The amplitudes of the GABA responses did not vary with the inhibitory or enhancing responses to pregabalin. The possibility of pregabalin-evoked inhibition of I Ca taking place through activation of G-protein coupled GABA B receptors, was then assessed. CGP52432 (10 μM) when applied alone had no effect on I Ca . In the presence of the GABA B receptor antagonist CGP52432 (10 μM), pregabalin (2.5 μM) evoked a 27 ± 3 % (n = 9) inhibition of the voltage-activated Ca 2+ current and complete recovery was observed 5 minutes after removal of the drug-containing pipette (Figure 8A ). In Ca 2+ imaging experiments 10 μM CGP52432 had no effect on the responses to 25 μM pregabalin. In the presence of CGP52432, 4 out of 16 neurones showed enhanced K + -evoked Ca 2+ influx (to 161 ± 24 % (n = 4) of control) in response to pregabalin and 12 neurones showed inhibition of Ca 2+ influx (64 ± 8 % (n = 12) of control) in response to pregabalin (Figure 8B & 8C ). Figure 8 The GABA B receptor antagonist CGP52432 had no effect on the actions of pregabalin. A) Bar chart showing the inhibitory effects of 2.5 μM pregabalin in the presence of 10 μM CGP52432 (PGB & CGP) on Ca 2+ current amplitude measured at the peak inward current (Peak) and at the end of a 100 ms voltage step command to 0 mV (End). The inset traces show the inhibition of the Ca 2+ current produced by 3 minutes application of pregabalin and CGP52432 (PGB + CGP) and partial recovery 5 minutes after removal of the drug perfusion pipette. B & C) Show example records of 25 μM pregabalin in the presence of 10 μM CGP52432 (PGB + CGP) modulating Ca 2+ flux evoked by 30 mM KCl. The open bars show the period of stimulation with K + and the filled bar the application of pregabalin in the presence of CGP52432 (PGB +CGP). Is there a role for G-proteins or G-protein coupled receptors in the responses to pregabalin in cultured dorsal root ganglion neurones? Previously, we found that the actions of gabapentin on I Ca were attenuated by pre-treating DRG neurones with pertussis toxin, an effect that did not appear to involve metabotropic GABA B receptors [ 12 ]. In this study several different approaches were taken to examine the influence of receptor and G-protein function in pregabalin actions. Firstly, the effects of a novel thiadiazole compound, SCH-202676, which inhibits ligand binding to a variety of G-protein coupled receptors (opioid, adrenergic, muscarinic and dopaminergic) were investigated. The selective and reversible action of SCH-202676 appears to involve allosteric modulation of both agonist and antagonist binding to G-protein coupled receptors [ 21 ]. Secondly, the potential influences of intracellular flash photolysis of caged GTP-γ-S and subsequent G-protein activation on pregabalin responses were also examined. SCH-202676 (10 μM) was applied to the intracellular environment via the patch pipette solution. After 5 minutes equilibration the whole cell Ca 2+ current had a mean amplitude of -0.54 ± 0.08 nA (n= 7). During this period there was a clear reduction in the inward current, although a steady state current level was reached. However, subsequent application of pregabalin (2.5 μM) resulted in a further significant reduction in Ca 2+ current amplitude to -0.43 ± 0.07 nA (n = 7, p <0.05). This represents a mean inhibition of 22 ± 5 % by pregabalin, a value very similar to the percentage inhibition produced by pregabalin in the absence of SCH-202676 (21 ± 3%). Therefore SCH-202676 had no effect on the pregabalin responses, indicating that pregabalin was not acting through an SCH-202676-sensitive G-protein coupled receptor (data not shown). Preliminary Ca 2+ imaging experiments were also conducted to determine whether extracellular SCH-202676 (10 μM, continually applied throughout the experiment) altered the effects of pregabalin (2.5 μM) on K + -evoked Ca 2+ transients. Under these experimental conditions pregabalin was found to markedly enhance or inhibit K + -evoked Ca 2+ transients (data not shown). This data again suggests that pregabalin was not acting through mechanisms that were sensitive to SCH-202676. Caged GTP-γ-S (100 μM) was applied to the intracellular environment via the patch pipette solution and after entering the whole cell recording configuration DRG neurones were left for 5 minutes to equilibrate. Once a stable control Ca 2+ current was obtained, three 200 V flashes of intense near UV light were applied to the neurone to achieve intracellular flash photolysis of the caged GTP-γ-S. We estimate that approximately 15 μM GTP-γ-S was photoreleased by the three flashes. As previously observed intracellular flash photolysis of caged GTP-γ-S reduced the amplitude and slowed the activation of I Ca [ 22 ]. The effects of photoreleased GTP-γ-S were attenuated by applying a large depolarising pre-pulse [ 23 ], which is consistent with voltage-dependent G-protein modulation of Ca 2+ channels (Figure 9A ). No recovery from photoreleased GTP-γ-S was observed during the period of the experiment. GTP-γ-S and pregabalin (2.5 μM) had additive inhibitory effects on I Ca . These additive actions of GTP-γ-S and pregabalin were apparent regardless of the order of application. One set of experiments was performed in which GTP-γ-S was photoreleased in the intracellular environment and then after stabilisation of the response pregabalin was applied (Figure 9B ). In another set of experiments pregabalin was applied first and after equilibration GTP-γ-S was photoreleased in the continual presence of pregabalin (Figure 9C ). When pregabalin (2.5 μM) was applied first it produced a mean inhibition in I Ca by 25 ± 5 % (n = 4), when applied after photorelease of GTP-γ-S, pregabalin produced a 20 ± 10 % (n = 5) inhibition of I Ca . Figure 9 Intracellular photorelease of GTP-γ-S did not alter the sensitivity of DRG neurones to pregabalin. A) Traces showing a Ca 2+ current attenuated and slowed by intracellular flash photolysis of caged GTP-γ-S (GTP-γ-S) and a Ca 2+ current showing voltage-dependent partial recovery of GTP-γ-S-evoked inhibition (GTP-γ-S + PP {pre-pulse to +120 mV}). B) Bar chart showing control (Ctrl) data, the inhibitory effects of both intracellular photorelease of GTP-γ-S (GTP-γ-S) and subsequent application of 2.5 μM pregabalin (GTP-γ-S & PGB) on the mean peak Ca 2+ current amplitude. Inset records show individual Ca 2+ currents recorded under control conditions, after intracellular photorelease of GTP-γ-S (GTP-γ-S) and after subsequent extracellular application of pregabalin for 3 minutes (GTP-γ-S + PGB). C) Bar chart showing control (Ctrl) data, the inhibitory effects of both 3 minutes extracellular application of 2.5 μM pregabalin (PGB) and subsequent intracellular photorelease of GTP-γ-S in the continued presence of pregabalin (PGB & GTP-γ-S) on the mean peak Ca 2+ current amplitude. Inset records show individual Ca 2+ currents recorded under control conditions, after application of pregabalin for 3 minutes (PGB) and after intracellular photorelease of GTP-γ-S (PGB & GTP-γ-S). None of the currents in this figure have been leak subtracted. However, neither pregabalin nor GTP-γ-S altered the leak current. Actions of pregabalin on voltage-activated potassium currents It was clear from the Ca 2+ imaging experiments that not all the cellular actions of pregabalin could be explained by the inhibition of voltage-activated Ca 2+ channels because both inhibition and enhancement in K + evoked Ca 2+ flux was observed. To investigate the actions of pregabalin further, its' effects on voltage-activated K + currents in DRG neurones were studied. Previously, Stefani and colleagues found that gabapentin modulated neuronal steady state non-inactivating K + currents [ 18 ]. Three minutes application of pregabalin (2.5 μM) had no significant effect on the voltage-activated K + current activated from a holding potential of -90 mV by a 100 ms voltage step command to 0 mV (Figure 10A ). However, raising the pregabalin concentration to 250 μM resulted in significant modulation of K + currents in DRG neurones. Pregabalin (250 μM) applied for 3 to 5 minutes produced an enhanced K + current in 11 out of 21 neurones but a modest inhibition of the outward current in 10 out of 21 neurones (Figure 10B and 10C ). DRG neurones are a heterogenous population of neurones and there is evidence that they express at least 6 diverse voltage-activated K + channels and that this expression is in part determined by the type of DRG neurone [ 24 ]. To distinguish between the two responses experiments were carried out on DRG neurones held at -30 mV to inactivate a proportion of the outward current. Under these conditions pregabalin inhibited the outward current in 9 out of 11 DRG neurones but still produced enhancement in 2 neurones. The level of inhibition increased at a holding potential of -30 mV to 30 ± 7% (n = 9) compared to 15 ± 4 % (n = 10) at -90 mV, (data not shown). Figure 10 Pregabalin modulates voltage-activated potassium currents in cultured DRG neurones. A) Traces showing that 2.5 μM pregabalin failed to significantly alter the outward K + current evoked by a 100 ms voltage step command from -90 mV to 40 mV. B) Current / voltage relationship showing enhancement of outward current following 3 minutes application of 250 μM pregabalin (filled circles = control data and open circles currents recorded in the presence of pregabalin (PGB); * = P < 0.05 & ** = P < 0.01). Inset traces show the control outward K + current activated at +40 mV and the enhanced K + current recorded in the presence of pregabalin (250 μM). C) Current / voltage relationship showing inhibition of outward K + current following 3 minutes application of 250 μM pregabalin (filled circles = control data and open circles K + currents recorded in the presence of pregabalin (PGB); * = P < 0.05). Inset traces show the control outward K + current activated at +40 mV and the attenuated K + current recorded in the presence of pregabalin (250 μM). Interestingly, in apparent contrast to pregabalin, 250 μM gabapentin was initially only found to inhibit K + currents in DRG neurones (Figure 11A,11B ). However, when studies into long-term (10–15 minutes) actions of gabapentin were investigated slowly developing outward current enhancement was identified. So similar to actions of pregabalin, biphasic responses to gabapentin were found with initial inhibition of K + currents and then a delayed enhancement of the outward current (Figure 11C ), [ 5 ]. Figure 11 Acute application of gabapentin produced modest inhibition of voltage-activated K + currents in cultured DRG neurones but long-term measurement of K + currents shows a delayed enhancement in outward current. A) Bar chart showing data for the acute (3–5 minutes) reversible inhibition of the mean K + current by 250 μM gabapentin (GBP). B) Traces showing a control outward K + current activated at 0 mV and the inhibited current activated at the same voltage after 3 minutes application of 250 μM gabapentin (GBP). C) Traces showing the biphasic response to 250 μM gabapentin. Illustrated are the control current, the inhibited current recorded after 5 minutes application of gabapentin (GBP) and the enhanced outward current measured 5 minutes after removal of the perfusion pipette containing gabapentin (Enhanced Current). Long-term modulation of K + current by pregabalin was then investigated. Dramatic increases in outward current were observed after a delay. This effect of pregabalin persisted even as pregabalin was removed from the extracellular environment, which may implicate a metabolic or intracellular signalling event in this response. The mean K + current amplitude at +40 mV increased from 3.37 ± 0.73 nA to 7.56 ± 1.10 nA (n = 11; p <0.01) 15 minutes after perfusion of 250 μM pregabalin. Figures 12A and 12B show an individual example record and trace of the delayed response to pregabalin. These responses did reverse but this took about 40 minutes with the response developing 3–10 minutes after the start of pregabalin application. No change in holding current or in the leak conductances were associated with the long-term effect of pregabalin and in the absence of pregabalin stable K + currents were recorded from DRG neurones for 16 minutes (n = 10). Figure 12 Pregabalin produced delayed enhancement of K + currents in cultured DRG neurones. A) Line graph showing a time course for the delayed action of pregabalin in a single neurone, the open bar shows the period of application of pregabalin (250 μM). Little effect of pregabalin was seen in this neurone until 5 minutes after removal of the pressure ejection pipette containing pregabalin. B) Inset traces show a control K + current activated at +40 mV, modest enhancement of the K + current after 5 minutes application of 250 μM pregabalin and the enhanced outward K + current recorded 9 minutes after application of pregabalin. Apamin mimicked the inhibitory action of pregabalin on K + current but did not prevent enhancement of the K + current by acute application of 250 μM pregabalin. C) Bar chart showing the modest inhibitory effect of apamin (1 μM) on K + current and the enhancement in K + current when pregabalin was applied in the continued presence of apamin (Apamin & PGB). D) Traces showing a control K + current, the inhibited K + current in the presence of 1 μM apamin and the enhanced K + current observed when apamin and pregabalin were applied together. Apamin prevented the inhibitory effect of pregabalin but not the enhancement of K + current. Several pharmacological experimental approaches were taken to characterise the biphasic actions of pregabalin and to determine the possible mechanism of action associated with the long-term K + current modulation. Apamin, a toxin from the honeybee, was used to block small conductance Ca 2+ -activated K + currents. Apamin (1 μM) caused a reduction in outward current at +40 mV from 2.64 ± 0.33 nA to 2.37 ± 0.39 nA, subsequent application of 250 μM pregabalin enhanced the current to 3.02 ± 0.39 nA (n = 11; p <0.01). No inhibitory effects were seen with pregabalin after treatment with apamin, suggesting that it is the apamin-sensitive Ca 2+ -activated K + channels that are inhibited by pregabalin (Figure 12C & 12D ). Gabapentin can be transported into cells via the L α-amino acid transporter and may achieve intracellular concentrations 10–20 times the levels in the extracellular environment [ 25 ]. Furthermore, actions through intracellular signalling pathways and specifically protein kinase A, have been proposed for gabapentin [ 12 ]. To examine whether the long-term enhancement of K + currents by pregabalin might involve intracellular sites of action, we applied pregabalin to the intracellular environments of DRG neurones via the patch pipette solution. Intracellular pregabalin (250 μM) evoked an increase in the K + current that started to develop within the first minute of entering the whole cell recording configuration. After an initial increase that could reflect equilibration with the KCl-based patch pipette solution containing pregabalin, a slow sustained significant increase in the outward K + current continued to develop over a 15 minute period (n = 10; p < 0.01; Figure 13A,13B ). No similar change in K + current was observed in control studies carried out in the absence of pregabalin (Figure 13C,13D ). Figure 13 Intracellular application of pregabalin enhanced K + currents in cultured DRG neurones. For these experiments pregabalin was included in the KCl-based patch pipette solution at a concentration of 250 μM. A) Bar chart showing the mean amplitudes of the K + current recorded immediately after entering the whole cell recording configuration (Initial I K ) and the maximum outward current recorded within 16 minutes of entering the whole cell recording configuration (Max I K ). B) Traces of the first K + current recorded using a patch pipette solution containing 250 μM pregabalin (Initial) and from the same cell the maximum outward K + current recorded with intracellular pregabalin (Max). C) Bar chart showing control data recorded from neurones not exposed to pregabalin. Illustrated are the mean amplitudes of the K + current recorded immediately after entering the whole cell recording configuration (Initial I K ) and the maximum K + outward current recorded within 16 minutes of entering the whole cell recording configuration (Max I K ). D) Traces of the first K + current recorded using the standard KCl-based patch pipette solution (Initial) and from the same cell the maximum outward current recorded under control conditions, within 16 minutes of entering the whole cell recording configuration (Max). Pertussis toxin pre-treatment results in the uncoupling of sensitive G-proteins from effector mechanisms, including voltage-activated Ca 2+ and K + channels. Additionally, pertussis toxin pre-treatment has previously been found to influence gabapentin actions on I Ca . Pertussis toxin pre-treatment (500 ng/ml; 16–18 hours) prevented pregabalin-induced long-term K + current enhancement in cultured DRG neurones (Figure 14A & 14B ). Figure 14 Pertussis toxin pre-treatment and intracellular (Rp)-cAMP prevented enhancement of K + current by pregabalin. A) Bar chart showing the mean amplitude of K + current recorded from DRG neurones pre-treated with pertussis toxin for 16–18 hours with 500 ng/ml (PTX Control) and after application of 250 μM pregabalin (PTX PGB), long term, up to 15 minutes monitoring of the current. B) Traces showing outward K + currents recorded from a DRG neurone pre-treated with pertussis toxin, prior to pregabalin application (PTX Control) and 10 minutes after application of 250 μM pregabalin. C) Bar chart showing mean data obtained from neurones containing (Rp)-cAMP (30 μM), which was applied to the intracellular environment via the patch pipette solution. Data shows the mean K + current amplitude recorded 1 minute and 5 minutes after entering the whole cell recording configuration (1 min; 5 min), after 5 minutes application of 250 μM pregabalin (PGB) and 10 minutes after removal of the pressure ejection pipette containing pregabalin. D) Traces from a single experiment showing the outward K + currents at 1 and 5 minutes after entering the whole cell recording configuration and allowing entry of 30 μM (Rp)-cAMP in the DRG neurones. Also shown are the K + current inhibited by 5 minutes application of 250 μM pregabalin (PGB) and the recovery of the K + current after the pressure ejection pipette containing pregabalin was removed. Intracellular (Rp)-cAMP prevented the delayed long-term enhancement of the K + current evoked by pregabalin. The possible role of cAMP-dependent protein kinase A (PKA) in the pregabalin-induced enhancement of K + current was then assessed using the inhibitor (Rp)-cAMP [ 26 ]. Intracellular application of 30 μM (Rp)-cAMP applied via the KCl-based patch pipette solution had no effect over a 5 minute equilibration period on voltage-activated K + current (n = 9). This may indicate that PKA has little or no basal or tonic activity on K + currents in DRG neurones in culture. However, when pregabalin was applied to DRG neurones loaded with (Rp)-cAMP no long-term enhancement of the outward current was observed. Under these recording conditions pregabalin still produced some inhibition of the K + current (Figure 14C & 14D ). These data provide evidence that the long-term modulation of voltage-activated K + channels by pregabalin is dependent on PKA-mediated phosphorylation. Conclusions In conclusion, these results indicate that pregabalin acts via the same basic mechanisms as gabapentin to inhibit voltage-activated Ca 2+ channels and that these inhibitory actions are independent of GABA receptor activation. Some features of the actions of pregabalin on intermediate size and large DRG neurones appear not to be seen with gabapentin. However, these distinct responses involve enhanced K + -evoked Ca 2+ transients by pregabalin rather than inhibition of Ca 2+ channels. Alpha 2 δ subunits of voltage-activated Ca 2+ channels remain a possible site of action for both pregabalin and gabapentin. Dooley and colleagues showed that both gabapentin and pregabalin attenuated K + -evoked norepinephrine release from rat neocortical slices by inhibiting P/Q-type Ca 2+ channels [ 10 ]. In cortical pyramidal neurones gabapentin predominantly works through L-type Ca 2+ channels [ 27 ]. Our work with Bay K8644 indicates that in cultured DRG neurones gabapentin [ 5 ] and pregabalin act predominantly independently of L-type channels. A recent study has also indicated that acting via G-protein coupled GABA B receptors, gabapentin selectively inhibited N-type Ca 2+ channels in hippocampal pyramidal neurones [ 15 ]. Interestingly, this selectivity of gabapentin seen in the hippocampus is different from the Ca 2+ channel modulation seen with the GABA B receptor agonist, baclofen. Our previous investigation using "toxityping" showed that gabapentin inhibited a variety of Ca 2+ channels in DRG neurones [ 11 ]. We conclude from all this work that gabapentin and pregabalin may have allosteric interactions with promiscuous α 2 δ Ca 2+ channel subunits. These α 2 δ subunits are not specifically combined with distinct pore forming Ca 2+ channel subunits (α 1 ) in all neurones and may therefore inhibit pharmacologically diverse Ca 2+ channels depending on the expression of Ca 2+ channel subunits in different neurones. However, transfection studies have shown that oocytes expressing Ca 2+ channels (Ca v 2.2) containing β1b and α 2 δ-1 or α 2 δ-2 subunits are insensitive to acute application of 50 μM gabapentin [ 28 ]. This work raises the possibility that other mechanisms independent of α 2 δ – Ca 2+ channel subunits are involved in the modulation of neuronal excitability by gabapentin. Additionally, the modulation of voltage-activated Ca 2+ channels by gabapentin and pregabalin acting through indirect mechanisms has been suggested. Candidates for such indirect mechanisms include the control of Ca 2+ channel functional expression [ 29 , 30 ] and metabotropic mechanisms linked to pertussis toxin sensitivity and PKA activation [ 12 ]. In hippocampal pyramidal neurones, uncoupling G-proteins from metabotropic receptors with N-ethylmaleimide prevents modulation of K + and Ca 2+ channels by gabapentin [ 15 ]. However, in DRG neurones it is not clear how pertussis toxin influences the actions of gabapentin and pregabalin but it may involve disruption of multi-protein complexes of G-proteins and ion channel subunits after ADP-ribosylation of Gα. The role of any G-protein coupled receptors in either gabapentin or pregabalin responses in DRG neurones are not supported by our investigations with SCH-202676. Furthermore, receptor and direct G-protein involvement in pregabalin effects on Ca 2+ channels also appears unlikely in DRG neurones. This is because in this study we found that intracellular flash photolysis of GTP-γ-S had no influence on pregabalin-evoked current inhibition and pregabalin did not alter responses to GTP-γ-S. Our findings are in agreement with those made in a previous study on cells expressing GABA B1a/B2 or GABA B1b/B2 receptor subunits. In this study GABA and the GABA B receptor agonist baclofen evoked [ 35 S]-GTP-γ-S binding responses but even at high concentrations both gabapentin and pregabalin did not [ 31 ]. The experiments designed to assess potential roles of GABA receptors in the responses to pregabalin showed that GABA receptor desensitization or the blockade of GABA B receptors did not attenuate pregabalin actions. These findings add to the published studies that indicate that gabapentin and pregabalin effects are independent of GABA receptor activation at least in some preparations [ 31 , 32 ]. Specifically, in cultured DRG neurones gabapentin was previously found not to activate a Cl - conductance to alter membrane potential and input resistance and the GABA B receptor antagonist saclofen did not influence the inhibitory action of gabapentin on I Ca [ 12 ]. Pregabalin has a higher affinity for α 2 δ subunits than gabapentin and in a number of studies is more effective. When applied alone pregabalin produced enhancement in Ca 2+ flux in some neurones, an effect not seen with gabapentin. So, it was surprising to see in the imaging experiments that when gabapentin and pregabalin were applied together they produced only inhibitory effects. This may reflect the allosteric interactions between these drugs and α 2 δ subunits of Ca 2+ channels as well as indirect modulation of Ca 2+ dependent conductances and interactions with components of cell signalling. There appear to be some inconsistencies in the pregabalin data when its actions on voltage-activated Ca 2+ currents are compared with effects on K + -evoked Ca 2+ transients. This may in part be due to effects of pregabalin on membrane conductances in addition to voltage-activated Ca 2+ currents. These effects may not be detected under voltage clamp recording conditions where Na + and K + currents are blocked to isolate Ca 2+ currents. In the imaging experiments all voltage-activated conductances are intact and may be modulated by pregabalin. Alternatively, pregabalin could potentially have a direct or indirect influence on Ca 2+ -induced Ca 2+ release from intracellular stores and alter Ca 2+ homeostatic mechanisms. These effects may be modulated differently in different sub-populations of DRG neurones and so produce the mixed responses recorded in small, intermediate and large neurones in this study. Thus different actions of pregabalin may be detected using fura-2 fluorescence imaging but these additional mechanisms may not influence the measurements of Ca 2+ currents. Although mixed responses to gabapentin were not previously identified, neither were clear effects on the amplitude of K + -evoked Ca 2+ transients apparent in DRG neurones [ 11 ], although they were seen in differentiated F-11 cells [ 12 ]. These apparent anomalies seen with pregabalin do not appear to be due to drug effects on Ca 2+ -induced Ca 2+ release or Ca 2+ homeostatic mechanisms such as Na + /Ca 2+ exchange. This conclusion was reached because pregabalin did not influence Ca 2+ transients evoked by caffeine and still enhanced K + -evoked Ca 2+ transients in choline chloride-based (low Na + ) extracellular solution. The mixed responses to pregabalin seen in the Ca 2+ imaging experiments prompted us to investigate the actions of pregabalin on voltage-activated K + currents, which could be modulated to cause an increase in Ca 2+ influx. This speculation was supported by the finding that application of the K + channel inhibitor, TEA, enhanced K + -evoked Ca 2+ transients. Stefani and colleagues have reported inhibition of outward K + currents by gabapentin [ 18 ]. Additionally, in rat hippocampal and human neocortical brain slices gabapentin has been shown to inhibit K + -evoked [ 3 H]-noradrenaline release. The activation of K ATP channels is implicated in these responses because not only does glibenclamide, a K ATP channel antagonist, attenuate the gabapentin response but also pinacidil, a K ATP channel agonist, mimics the response and did not have additive effects with gabapentin [ 33 ]. Mixed or biphasic responses to gabapentin and pregabalin were seen, with both inhibition and enhancement of K + currents observed. The inhibitory actions of both gabapentin and pregabalin on K + currents recorded from DRG neurones may not reflect a direct action of these drugs on K + channels. The effects of apamin, which prevented the inhibition of outward current by pregabalin indicated that small conductance Ca 2+ -activated K + channels were involved in the response. This is consistent with the effects of gabapentin seen in isolated cortical neurones [ 18 ]. The pregabalin inhibitory action on voltage-activated Ca 2+ channels may underlie the inhibition of outward K + current. A reduction in Ca 2+ influx would lead to reduced activation of Ca 2+ -activated K + channels and therefore a smaller outward current. If this indirect effect of pregabalin resulted in a disproportionate reduction in Ca 2+ -activated K + conductances this may result in prolonged depolarisation. In turn a prolonged depolarisation may maintain activation of Ca 2+ channels that are insensitive to pregabalin and an increase in Ca 2+ influx as seen in a sub-population of larger DRG neurones in the Ca 2+ imaging experiments. It is not clear why gabapentin, which produced modest inhibition of K + currents, did not also produce enhancement of K + -evoked Ca 2+ transients. It may be explained by a balance between inhibition of Ca 2+ influx through voltage-activated channels and modulation of Ca 2+ flux through inhibition of K + conductance. Consistent with this hypothesis, gabapentin appears to be less effective than pregabalin at modulating K + -evoked Ca 2+ transients in DRG neurones. This is supported by our previous study that showed that gabapentin did not significantly alter the peak Ca 2+ transients [ 11 ]. The delayed enhancement of voltage-activated K + currents by gabapentin and pregabalin is a powerful mechanism for reducing cell excitability. The slow development of this response and its gradual decline suggested that intracellular signalling events were involved. Pregabalin produces enhancement of K + currents when applied either inside or outside DRG neurones. The delay in the development of these responses is less with intracellular application of pregabalin, suggesting that it may have an intracellular site of action. This effect may also depend on drug uptake into cells via L α-amino acid transporters and the presence and activity of these transporters may greatly influence cell sensitivity to gabapentin and pregabalin. Although it is not clear how gabapentin and pregabalin might activate PKA, it is a candidate target site. Our previous work showed that inhibition of Ca 2+ currents by gabapentin was sensitive to cAMP analogues that activated or inhibited PKA [ 12 ]. Similarly, in this present study (Rp)-cAMP blocked enhancement of K + currents by pregabalin. In the literature there are a number of reports of PKA altering neuronal excitability and modulating K + conductances. PKA is involved in pain signalling, playing a role in prostaglandin-induced activation and sensitization of DRG neurones [ 34 ]. PKA activity attenuates KV3.2 channel currents [ 35 ] and A-type current [ 36 ] and is involved in the inhibition of K + conductances by prostaglandin E-2 [ 37 ]. Our results are not explained by these findings but PKA has also been implicated in presynaptic inhibition of GABA release [ 38 ], large conductance calcium – and voltage-activated K + channel activity [ 39 ] and cannabinoid receptor mediated modulation of ion channels [ 40 ]. In the context of our project, of particular interest is the biphasic modulation in retinal cones of voltage dependent K + and Ca 2+ channels by a synthetic cannabinoid receptor agonist and the sensitivity of these responses to Wiptide, a PKA inhibitor [ 40 ]. However, the G-protein pharmacology is different in the cone cells suggesting different PKA activation pathways in DRG neurones exposed to pregabalin. There appears to be a considerable number of apparently conflicting effects mediated by PKA, although a number of different preparations have been studied. The concept of conditional protein phosphorylation such that initial phosphorylation state affects the sensitivity of different effector targets to subsequent activation [ 39 ] could determine responses to gabapentin and pregabalin acting through PKA. A future challenge will be to determine how PKA plays roles in activation and sensitization of DRG neurones and also plays a role in gabapentin and pregabalin responses to dampen down electrical excitability in the same neurones. Perhaps answers will come when we have a better understanding of the changes in neuronal phenotype that develop with pain disorders? An interesting feature of some actions of gabapentin and pregabalin is that their effects often outlast the period of drug delivery. Maintained intracellular signalling effects of these drugs and the delayed responses involving PKA modulation of Ca 2+ and K + conductances may underlie this characteristic. Gabapentin has previously been shown to attenuate high frequency action potential firing both in CNS neurones [ 41 ] and in cultured DRG neurones [ 5 , 12 ]. Pregabalin produces the same effect, dramatically but reversibly reducing the number of action potentials fired in response to 300 ms depolarising current commands. The underlying mechanisms involved in these reductions in electrical excitability could involve inhibition of voltage-activated Ca 2+ channels with resulting reduced activation of Ca 2+ -activated K+ currents and effects on voltage-dependent ion channel availability. Although the delayed enhancement of voltage-activated K + channels could contribute, this effect appears to develop over a longer period. These actions may contribute both to anticonvulsant effects of gabapentin and pregabalin as well as their therapeutic actions in pain disorders. It is particularly worth noting that the inhibition of Ca 2+ influx is most consistently seen in small diameter cultured DRG neurones that are likely to be the pain fibres. Our data adds to previous work that has assessed the actions of gabapentin and pregabalin on K + -evoked neurotransmitter release in CNS neurones [ 10 ] and the pharmacology of pain in whole animal models [ 8 , 9 , 42 - 44 ]. The present work and these previous studies suggest that gabapentin and pregabalin mostly act through the same basic mechanisms. These mechanisms in cultured DRG neurones appear to involve allosteric inhibitory modulation of Ca 2+ channels through drug interactions with α 2 δ subunits, reduced activation of Ca 2+ -dependent conductances and modulation of both Ca 2+ and K + channels through metabotropic mechanisms that involve PKA. Interestingly, both gabapentin and pregabalin inhibit release of sensory peptides (substance P and CGRP) this may provide a mechanism of action down-stream of ion channel modulation. This effect is only observed after inflammation and may contribute to their analgesic properties of gabapentin and pregabalin [ 45 ]. Ca 2+ channels in DRG neurones are targets for antinociceptive agents but the pharmacology of these channels remains to be fully exploited in pain therapy [ 46 , 47 ]. Further studies on changes in ion channel subunit expression and alterations in intracellular signalling pathways in pain disorders may in the future open up new therapeutic opportunities. Methods Cell culture One to four-day old Sprague-Dawley rats were decapitated and dorsal root ganglia removed. DRG neurones were dissociated enzymatically (0.125% collagenase for 13 minutes and 0.25% trypsin for 6 minutes) and mechanically (trituration). Primary cultures of DRG neurones were plated on lamin-polyornithine coated coverslips and bathed in Ham's F-14 culture medium (Imperial Laboratories) containing 10% horse serum (Gibco), low NGF (20 ng/ml; Sigma), NaHCO 3 (14 mM), streptomycin (50 μg/ml) and penicillin (50 IU/ml). The cultures were maintained for up to two weeks at 37°C in humidified air with 5% CO 2 . Cultures were re-fed with fresh media after 5 days. In some experiments DRG neurones were pre-treated with pertussis toxin (500 ng/ml; for 18 hours) to ADP-ribosylate the α subunits of certain G-proteins. This prevents pertussis toxin-sensitive G-protein being activated through a range of G-protein coupled receptors and inhibits coupling to voltage-activated Ca 2+ channels and other potential effectors [ 12 , 48 ]. Electrophysiology and calcium imaging The whole cell patch clamp recording method and fura-2 Ca 2+ imaging were used to measure the inhibitory actions of pregabalin and gabapentin on Ca 2+ entry through voltage-activated channels. Initially, multiple firing properties of a sub-population of DRG neurones were studied using a patch pipette solution containing in mM: KCl, 140; EGTA, 5; CaCl 2 , 0.1; MgCl2, 2.0; HEPES, 10.0; ATP, 2.0. The extracellular solution containing in mM: NaCl, 130; KCl, 3.0; CaCl 2 2.0; MgCl 2 , 0.6; NaHCO 3 1.0, HEPES 10.0 and glucose 5.0. For recording Ca 2+ channel currents the patch pipettes were filled with CsCl-based solution containing in mM: 140 CsCl, 0.1 CaCl2, 5 EGTA, 2 MgCl2, 2 ATP, 10 Hepes. The pH and osmolarity of the patch pipette solutions were corrected to 7.2 and 310–320 mOsm.l -1 with Tris and sucrose. The extracellular bathing solution used to study Ca 2+ currents contained in mM: 130 choline chloride, 2 CaCl2, 3 KCl, 0.6 MgCl2, 1 NaHCO3, 10 HEPES, 5 glucose, 25 tetrethylammonium chloride, 0.0025 tetrodotoxin (Sigma). The pH and osmolarity of this extracellular bathing solution was corrected to 7.4 and 320 mOsml -1 with NaOH and sucrose respectively. The recording solutions used in these experiments were designed to attenuate voltage-activated Na + and K + currents and isolate voltage-activated Ca 2+ currents. The range of values for the series resistance under our recording conditions was from ~8 to 15 MΩ. Pregabalin and other drugs were applied to the extracellular environment by low-pressure ejection from a blunt pipette positioned about 50–100 μm away from the cell being recorded. Voltage-activated Ca 2+ currents were evoked by 100 ms voltage step commands applied every 30 s. Similar protocols were used to study the actions of pregabalin on voltage-activated K + currents but NaCl-based extracellular bathing solution and KCl-based patch pipette solutions were used. In some experiments GTP-γ-S was photoreleased inside DRG neurones. Caged GTP-γ-S (100 μM; Molecular Probes) was included in the CsCl-based patch pipette solution. After obtaining control Ca 2+ currents intracellular flash photolysis was achieved by flashing the neurone being studied (three 200 V flashes to produce ~15μM GTP-γ-S) using a XF-10 xenon flash lamp with a UG11 bandpass filter (Hi-Tech Scientific; [ 49 ]) All voltage-activated Ca 2+ and K + currents had scaled linear leakage and capacitance currents subtracted to obtain values for the net inward Ca 2+ current or net outward K + current. Data are given as mean ± standard error of the mean (s.e.m.) values and statistical significance was determined using a paired or independent Student's t test as appropriate. For Ca 2+ imaging in cultured DRG neurones the cultures were incubated for 1 hour in NaCl-based extracellular solution contained (in mM): NaCl, 130; KCl, 3.0; MgCl 2 , 0.6; CaCl 2 , 2.0; NaHCO 3 , 1.0; HEPES, 10.0; glucose, 5.0 and fura-2AM, 0.01; (Sigma, 1 mM stock in dimethylformamide). The pH was adjusted with NaOH to 7.4 and the osmolarity to 310–320 mOsm with sucrose. The cells were then washed for 10–20 minutes with NaCl-based extracellular solution to remove the extracellular fura-2AM and this period allowed cytoplasmic de-esterification of the Ca 2+ sensitive fluorescent dye. The cells were constantly perfused with NaCl-based extracellular solution (1–2 ml/min) and viewed under an inverted Olympus BX50WI microscope with a KAI-1001 S/N 5B7890-4201 Olympus camera attached. Some Ca 2+ imaging experiments were carried out using a Ca 2+ -free NaCl-based solution (as standard NaCl-based extracellular solution but with no added CaCl 2 ) or choline chloride-based solution (as standard NaCl-based extracellular solution but with choline chloride in place of NaCl). The fluorescence ratiometric images from data obtained at excitation wavelengths of 340 nm and 380 nm were viewed and analysed using OraCal pro, Merlin morphometry temporal mode (Life Sciences resources, version 1.20). The DRG neurones were stimulated with NaCl-based extracellular solution containing high K + (30 mM), which produced depolarisation, activation of voltage-gated Ca 2+ channels and large transient increases in intracellular Ca 2+ . Three consistent transient increases in intracellular Ca 2+ could be obtained in a single experiment on cultured DRG neurones [ 11 ]. The actions of pregabalin and gabapentin (2.5–250 μM) were investigated on the response to the second stimulus in DRG neurones. The actions of pregabalin and gabapentin on the Ca 2+ transient amplitude, duration at 1/2 peak amplitude and total Ca 2+ flux were measured. All experiments were conducted at room temperature and data are expressed as means ± s.e.m. List of abbreviations DRG, Dorsal root ganglion. cAMP, Cyclic adenosine monophosphate GABA, Gamma aminobutyric acid GBP, Gabapentin GTP-γ-S, Guanosine 5'-o(3-thio)triphosphate I Ca , Calcium current NGF, Nerve growth factor PGB, Pregabalin PKA, Protein kinase A PP, Pre-pulse (Rp)-cAMP, (R)-adenosine, cyclic 3', 5'-(hydrogenphosphorothioate) triethylammonium Authors' contributions All the authors of this manuscript contributed to electrophysiological and Ca 2+ imaging experiments. DM and RHS designed and conducted the experiments on K + currents and wrote the first draft of this manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC514605.xml
521494	Open Access gains attention in scholar communication	null	Open Access is one of the attempts to maximize the exchange of information, and therefore benefits the scholar communication [ 1 ]. Molecular Cancer offers Open Access to all of its content, thereby providing a platform to present information to specialists and the public in order to further promote free exchange of ideas, concepts and findings in all fields of cancer-related biomedical science. All the published articles in the journal are determined by the peer review process. Open Access has following broad benefits for science and the general public: • All articles become freely and universally accessible online; so an author's work can be read by anyone at no cost. • The authors hold copyright for their work and grant anyone the right to reproduce and disseminate the article, provided that it is correctly cited. • A copy of the full text of each Open Access article is permanently archived in an online repository separate from the journal, such as PubMed Central, the US National Library of Medicine's full-text repository of life science literature, the repositories at the University of Potsdam in Germany, at INIST in France and in e-Depot, the National Library of the Netherlands' digital archive of all electronic publications. • Authors are assured that their work is disseminated to the widest possible audience. This is accentuated by the authors being free to reproduce and distribute their work, for example by placing it on their institution's website. It has been suggested that free online articles are more highly cited because of their easier availability [ 2 ]. • The information available to researchers will not be limited by their library's budget, and the widespread availability of articles will enhance literature searching. • The results of publicly funded research will be accessible to all interested readers and not just those with access to a library with a subscription. As such, Open Access could help to increase public interest in, and support of, research. Please note that this public accessibility may become a legal requirement in the USA if the proposed Public Access to Science Act is made law [ 3 ]. • A country's economy will not influence its scientists' ability to access articles because resource-poor countries (and institutions) will be able to read the same material as wealthier ones, although creating access to the internet is another matter. Molecular Cancer published a number of interesting papers, and the list of the top ten most accessed articles is available at . All papers accepted by Molecular Cancer appear as 'accepted manuscript' on the web pages and are subsequently included in PubMed. A fully formatted portable document file is available approximately two to three weeks after acceptance along with a web-version of the article. The on-line publication, to the exclusion of print, has many advantages: Coloured pictures can be presented along with large sets of supporting data (movies, tables, pictures, et cetera) without additional charges. In addition, the on-line submission process allows a fast and effective handling of papers and allows authors to check the status of their submitted manuscript(s). There is no limitation in space, but concise papers are more likely to be read. The peer review policy, described in [ 4 ], ensures a fair evaluation of the work. We wish to thank our authors for sending their work to Molecular Cancer , all members of the editorial board and the reviewers for their ongoing support for Open Access publishing and for aiming higher standards for Molecular Cancer . The acceptance rate of Molecular Cancer did not change significantly, compared to the last report [ 4 ]. One out of three incoming articles are accepted for publication at Molecular Cance r after revisions. In addition to indexing in PubMed, PubMed Central and other search engines, Molecular Cancer is working closely with the Institute for Scientific Information to ensure that citation analysis of our articles will be available. Competing interests PJC is Editor-in-Chief and CS is Deputy Editor of this journal. Both do not receive any remuneration for their efforts but they are exempted from the article processing fee for this journal.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC521494.xml
524249	Representation of Attended Versus Remembered Locations in Prefrontal Cortex	A great deal of research on the prefrontal cortex (PF), especially in nonhuman primates, has focused on the theory that it functions predominantly in the maintenance of short-term memories, and neurophysiologists have often interpreted PF's delay-period activity in the context of this theory. Neuroimaging results, however, suggest that PF's function extends beyond the maintenance of memories to include aspects of attention, such as the monitoring and selection of information. To explore alternative interpretations of PF's delay-period activity, we investigated the discharge rates of single PF neurons as monkeys attended to a stimulus marking one location while remembering a different, unmarked location. Both locations served as potential targets of a saccadic eye movement. Although the task made intensive demands on short-term memory, the largest proportion of PF neurons represented attended locations, not remembered ones. The present findings show that short-term memory functions cannot account for all, or even most, delay-period activity in the part of PF explored. Instead, PF's delay-period activity probably contributes more to the process of attentional selection.	Introduction Jacobsen (1935 , 1936 ) first discovered that damage to the primate prefrontal cortex (PF) appeared to cause a short-term memory deficit. In his experiments, monkeys and chimpanzees with bilateral damage to PF failed to retrieve food from one of two opaque cups when the food had been out of sight for even a few seconds. Intact animals could find the food 5 min or more after they had last seen it. Pribram et al. (1952) later identified the part of PF responsible for this deficit as area 46, also known as the dorsolateral prefrontal cortex (PFdl). More recently, temporary inactivations of portions of PFdl caused what appeared to be a short-term memory loss in localized regions of space ( Funahashi et al. 1993a ). Once the concept of working memory ( Baddeley 1986 ) became established in contemporary neuroscience (see Postle et al. 2003 ), these neuropsychological findings contributed to the theory that PF functions in working memory ( Goldman-Rakic 1987 ) and, in some extreme formulations, only in working memory. In the 1990s this theory developed a wide following, and the idea that PFdl functions in spatial working memory, with other parts of PF functioning in different kinds of working memory, became the predominant theory of PF function, especially for nonhuman primates. As important, the concept of working memory used by proponents of this theory focused mostly on the short-term maintenance of information, and rather less on the manipulation or monitoring of such information or on the use of that information for decisions. Accordingly, we refer to the former aspect of working memory as maintenance memory to distinguish it from the broader concept and do not use the phrase working memory elsewhere in this report. Note, however, that when we use the phrase maintenance memory, many authorities would use “working memory” instead. Consistent with the idea that PF functions predominantly in maintenance memory, delay-period activity in PF has often been interpreted as a memory trace (e.g., Funahashi et al. 1989 ; Romo et al. 1999 ; Constantinidis et al. 2001 ). The phrase delay-period activity applies to neuronal activity that follows the transient presentation of an instruction cue and persists until a subsequent “go” or “trigger” signal. The description of delay-period activity in PFdl appeared very early in the history of behavioral neurophysiology ( Fuster and Alexander 1971 ; Kubota and Niki 1971 ; Fuster 1973 ), and, in accord with the maintenance-memory theory, some PF cells appear to buffer activity representing remembered information, even when distracting stimuli appear during the delay period ( di Pellegrino and Wise 1993b ; Miller et al. 1996 ; Moody et al. 1998 ). Although the interpretation of delay-period activity in terms of the short-term memory of a stimulus has a long history, many studies have explored alternatives. Neurophysiological experiments designed to explore alternatives to the maintenance-memory interpretation of delay-period activity first attempted to dissociate sensory from motor signals. These studies showed that PFdl neurons preferentially reflected sensory signals, which supported the idea that these neurons encode stimulus memory over the short term. For example, one influential study used the “antisaccade” task ( Funahashi et al. 1993b ), in which a stimulus in one direction (from a central fixation point) instructed an eye movement in the opposite direction. More than twice as many PFdl neurons represented the location of the sensory stimulus as represented the target (or direction) of movement. In another experiment, when a given spatial cue guided two different reaching movements, motor factors affected PFdl neurons only rarely and weakly compared to neurons in the premotor cortex ( di Pellegrino and Wise 1993b ), especially when viewed at a population level ( Wise et al. 1996a ). These results supported the idea that more delay-period activity in PFdl reflected the memory of sensory cues than represented motor preparation or movement targets, but did not explore other alternative interpretations of delay-period activity. Neuroimaging studies have provided support for some of these alternatives. At first, neuroimaging studies appeared to back the maintenance-memory theory of PF function, which bolstered the interpretation of PF's delay-period activity in the context of that theory. After an initial period of nearly uniform support, however, subsequent neuroimaging studies have suggested that PFdl plays a role in aspects of attention and other functions instead of, or in addition to, maintenance memory. Indeed, one recent report disputed whether PF plays any role in short-term memory at all. To quote the investigators, “no part of frontal cortex, including PF, stores mnemonic representation[s] . . . reliably across distracted delay periods. Rather, working memory storage . . . is mediated by a domain-specific network in posterior cortex” ( Postle et al. 2003 ). Passingham and his colleagues have used the phrases attention to action, attention to intention, and attentional selection to describe certain PFdl functions ( Rowe et al. 2000 ; Rowe and Passingham 2001 ). Petrides and his colleagues have, likewise, emphasized a role for PFdl in monitoring items in memory ( Owen et al. 1996 ; Petrides et al. 2002 ). These alternative views of PF function point to a role in top-down control of attention and are supported by other neuroimaging and neuropsychological findings implicating PF in attentional functions (see Discussion). In sum, then, neuroimaging and neuropsychological findings bring into question the interpretation of PFdl's delay-period activity mainly in terms of maintenance memory. Previous neurophysiological experiments have ruled out motor factors, such as motor planning and the representation of the targets of movement, for most of PFdl's delay-period activity, but have typically lacked control over spatial attention. The present experiment tested an alternative to the maintenance-memory interpretation of PFdl's delay-period activity by pitting the representation of a remembered location against the representation of an attended location, when either location could serve as the target of an upcoming saccadic eye movement. Results Two monkeys performed the task depicted in Figure 1 A. Briefly, the monkeys maintained fixation on a spot presented at the center of a video screen, called the fixation point. A solid gray circle then appeared at a fixed distance from the fixation point in any one of the four cardinal directions ( Figure 1 A, part a): left, right, up, or down from center. Next, as central fixation continued, the gray circle revolved clockwise or counterclockwise around the fixation point, moving along a circular trajectory (arrow in Figure 1 A, part b). It then stopped at one of the four cardinal directions from center, after having revolved 90°, 180°, 270°, or 360° ( Figure 1 A, part b). After a variable delay period of 1.0–2.5 s, the circle brightened or dimmed for 150 ms ( Figure 1 A, part c) and then disappeared ( Figure 1 A, part d). The change in the circle's brightness served as the trigger signal for a saccadic eye movement (arrows in Figure 1 A, part d). On control trials, the circle either did not move or revolved 360° and stopped at its initial location for that trial. During those trials, both dimming and brightening of the circle instructed a saccade toward its location. During other trials, dimming and brightening of the circle guided both the timing of the response and the choice between two alternative saccade targets. Figure 1 Task and Behavior Behavioral task (A) and representative horizontal and vertical eye position records (B). (A) Each trial began when the monkeys pressed a button to make a fixation point (FP) appear at the center of the video monitor. Some time after the monkeys fixated the FP (dashed lines), a gray circle (depicted here as white) appeared at one of four peripheral locations. The figure illustrates its appearance at the 0° location (part a). The monkeys had to remember this location later in the task; hence we termed it the remembered location. On most trials, the circle subsequently revolved around the FP to a different location, as the monkeys maintained central fixation. The figure illustrates its termination at the 90° location (part b). A small change in the circle's luminance (part c) signaled the monkeys where to look next. This cue persisted for 150 ms, then disappeared. Because the monkeys depended on this subtle and brief cue for both timing and targeting information, we termed this the attended location. If the circle dimmed (dark gray, part c, top), the monkeys had to make a saccade to the attended location (Att-trials, part d, top). If the circle brightened (starburst, part c, bottom), the monkeys had to make a saccade to the remembered location (Rem-trials, part d, bottom). After saccade initiation, the central FP disappeared and, if the monkeys made a saccade to the correct location, a new FP appeared there (not shown). The monkeys had to fixate the new FP and, after it dimmed, release the button to produce a fruit juice reward. (Monkey drawing courtesy of Dr. Michael Shadlen.) Brightening of the circle indicated that the monkeys should make a saccade to the circle's initial location on that trial, which the monkeys had to remember in order to perform the task correctly ( Figure 1 A, parts c and d, bottom). Accordingly, we called these trials remembered-location trials (Rem-trials). Dimming of the circle signaled that the monkeys should make an eye movement to its current location ( Figure 1 A, parts c and d, top). We called these trials attended-location trials (Att-trials), for the following reasons. As a key feature of the experimental design, the circle's brightness changed only subtly and remained visible in its new form only briefly. Because the monkeys could not predict whether the circle would brighten or dim and because that subtle, short-lived event provided essential information about the time and target of the response, the monkeys had to attend to the circle intently during the period preceding the trigger signal. As a result of the central fixation requirement, this attention was necessarily covert, although it seems likely that the monkeys would have attended overtly to the circle (i.e., fixated it), had they been allowed to do so. Indeed, the monkeys did so during training. The Discussion takes up the issues of divided attention, multiple motor plans, default motor plans, and other interpretational issues. By varying the final location of the circle from trial to trial, we could test for significant spatial tuning for attended locations, and by varying the initial location of the circle, we could test for significant spatial tuning for remembered locations. In addition, we tested the monkeys' performance in a “no-memory” condition, which had the same the sequence of events as in the standard version of the task. In the “no-memory” condition, however, the initial location of the circle remained marked by a stationary stimulus identical to the circle that revolved around the fixation point. Behavior Figure 1 B shows selected eye-position records, matched to the trials illustrated in Figure 1 A. Table 1 shows that both monkeys achieved a high level of performance on this challenging task. For Rem-trials, these data show that the monkeys remembered the circle's initial location, and—because they could not know the trial type in advance of the trigger signal—they must have also done so for Att-trials. Table 1 Task Performance and Reaction Times for Each Monkey The percentage of correctly executed trials comes from the trials on which the monkey maintained fixation until the trigger signal occurred and then performed a saccade to the instructed (correct) or some other (incorrect) location. The reaction times come from correct trials only. Means (± SEM) are presented for different angular differences between the remembered and attended locations (0°, 90°, or 180°). For control trials, which correspond to a 0° difference (360° revolutions excluded), Att-trials are trials on which the circle dimmed, and Rem-trials are those on which the circle brightened Table 1 also shows the reaction times for each monkey. Taking the two monkeys together, saccades to the remembered location began approximately 36 ms later than those to the attended location, a difference that was highly significant (Wilcoxon rank sum test, p < 0.001). We can only speculate about the cause of this difference, but reaction times on Rem-trials may have been longer because attention had to be disengaged from the circle's location and reoriented to the remembered one prior to the response. For the “no-memory” condition (not given in Table 1 ), reaction times for Att-trials increased approximately 16 ms compared to the standard version of the task, whereas reaction times for Rem-trials decreased approximately 22 ms (both highly significant differences, Wilcoxon rank sum test, p < 0.001). These data are consistent with the idea that each of the two marked locations attracted attention in the no-memory condition, whereas the monkeys directed most of their covert attentional resources to the attended location in the standard version of the task. We acknowledge, however, that there are other interpretations of these data. On control trials, for example, when the saccade was always toward the circle, saccade initiation was approximately 18 ms slower when the circle brightened (as it did on Rem-trials) than on trials when it dimmed (as it did on Att-trials). Thus, factors other than the orientation of attention probably contributed to reaction-time differences. Single-Neuron Analysis Figure 2 illustrates the activity of a neuron tuned to the attended location during the delay period. Only activity collected during correctly executed trials appears in any of the analyses presented in this report. The figure shows histogram and raster displays of neuronal activity aligned on the trigger signal for Att-trials ( Figure 2 A) and Rem-trials ( Figure 2 B), arranged in the form of a matrix, as illustrated and labeled in Figure 2 C. Delay-period activity, enclosed by the red rectangles in Figures 2 A and 2 B, varied with the attended location (columns), but not with the remembered location (rows). The firing rate during the delay period was highest when the monkey attended to the 90° location (up from screen center, see Figure 1 A, part b). We called this the cell's preferred location. The lowest firing rate occurred when the monkey attended to the 270° location, termed the least preferred location. Figure 2 Example Neuron Representing the Attended Location In (A–C), the four rows correspond to different remembered locations and the four columns to different attended locations (see key in [C]). (A and B) PETHs and raster displays aligned on the trigger signal (vertical line). In the rasters, each dot represents a neuronal spike, and each line of dots shows a sequence of spikes during a single behavioral trial. (A) Trials in which the stimulus dimmed and the monkey made a saccade to the attended location (Att-trials). (B) Trials in which the stimulus brightened and the monkey made a saccade to the remembered location (Rem-trials). The activity of this neuron depended on where the monkey attended, with a preferred location of 90°. Note the large variation in firing rate from column to column (across the attended locations) and relative constancy of rate within columns (across remembered locations). (C) Compact representation of spatial tuning pattern shown in (A) and (B), combined. Each circle's area is proportional to the average firing rate during the 800-ms period immediately preceding the trigger signal (red rectangle in [A] and [B]). Note that the major diagonal of this firing-rate matrix, running from the upper left to the lower right corner, corresponds to the control trials, which lacked a memory requirement. F, maximal firing rate; sp/s, spikes per second. For each neuron, we assessed the extent of spatial tuning for the attended location with an index called attended-location index (I Att ), which measured the variability in discharge rate among attended locations. We assessed the extent of spatial tuning for the remembered locations with a related index called remembered-location index (I Rem ) (see Materials and Methods ). A neuron was considered spatially tuned if I Att , I Rem , or both significantly exceeded 1.0 (randomization test, p < 0.01; see Materials and Methods ). We classified neurons as attention cells if I Att attained statistical significance but I Rem did not, as memory cells for the opposite result, and as hybrid cells if both indexes showed statistical significance. Figure 3 A– 3 C shows examples of an attention cell, a memory cell, and two hybrid cells. ( Figures S1–S3 show the trial-by-trial activity for each of these four cells, both before and after the trigger signal.) Neurons tuned to the attended location (attention cells) dominated the neuronal sample in both monkeys, comprising 61% of cells spatially tuned during the pretrigger delay period ( Table 2 ). Neurons tuned to the remembered location (memory cells) made up 16% of the spatially tuned neurons, and those tuned to both locations (hybrid cells) amounted to 23%. For 27% of the hybrid cells, the attended and remembered locations associated with the highest firing rate were the same ( Figure 3 C, part a); in the remaining 73% of the hybrid cells, these preferred locations differed ( Figure 3 C, part b). Figure 3 Example Firing Rate Matrices and a Scatter Plot of Tuning Indexes PFdl neurons with different classes of spatial tuning. Firing rate matrices (A–C) in the format of Figure 2 C; (E) gives the key. Tuning selectivity indexes (I Att and I Rem ) and firing rate scale (F) appear adjacent to each firing rate matrix. (A) A neuron tuned to the attended location (different from the cell shown in Figure 2 ). (B) A neuron tuned to the remembered location. Its firing rate primarily varied across rows. (C) Two cells tuned to both the attended and remembered locations (hybrid neurons). One neuron (part a) exhibited a high firing rate when either the attended or remembered location was at 270°. The other neuron (part b) showed its highest activity when the remembered location was at 180°, but was inhibited when that was the attended location. (D) Scatter plot of spatial tuning indexes for attended (I Att ) and remembered (I Rem ) locations for each spatially tuned neuron in both monkeys. The different neuronal classes are color coded as in (A–C): blue corresponds to attention cells, red to memory cells, and green to hybrid cells. Table 2 Cell Classification Number of neurons that significantly ( p < 0.01) encoded the attended location (Attention), the remembered location (Memory), or both locations (Hybrid) during the 800 ms immediately prior to the trigger signal Figure 3 D illustrates the degree of tuning for both the attended (I Att ) and remembered ( I Rem ) locations. Each data point on the scatter plot represents a single spatially tuned neuron (both monkeys combined). Tuning for the remembered location (red symbols) was both weaker and less frequent than tuning for the attended location (blue symbols). Note that hybrid cells (green symbols) fill most of the space between the other two classes and that relatively few cells represent a single location exclusively. For example, many of the neurons classed as memory cells show some sensitivity to the attended location, albeit not a statistically significant one by the test that we employed. For the entire group of spatially tuned neurons ( n = 303, both monkeys and all three cell classes combined), the mean selectivity indexes (± SEM) for the attended and remembered locations were I Att = 1.84 ± 0.08 (median = 1.39, interquartile range [IQR] = 0.73) and I Rem = 1.21 ± 0.02 (median = 1.08, IQR = 0.23), which differed significantly at the p < 0.001 level (Wilcoxon matched-pairs test). Table 3 shows comparable data for each cell class and Figure S4 gives similar data for various combinations of these classes. The selectivity for the attended location also exceeded that for the remembered one when expressed in terms of firing rates. For the attended location, the difference in firing rate between the preferred and least preferred locations averaged 8.8 ± 0.5 spikes/s, which was significantly greater than the 5.3 ± 0.3 spikes/s for the remembered location (Wilcoxon matched-pairs test, p < 0.001). Table 3 Spatial Tuning Indexes Early Versus Late in the Trial Tuning indexes (mean ± SEM) were calculated from both the 800 ms immediately preceding circle movement (Early, I ) and the 800 ms immediately preceding the trigger signal (Late, I Att , I Rem ). For both attention and hybrid cells, spatial tuning to the attended location was significantly stronger (Wilcoxon matched-pairs test) late in the trial, when the monkeys awaited the trigger signal. Values for memory tuning (I Rem ) appear for completeness, not for statistical testing. See also Figure S4 We examined whether these results merely reflected the presence of a stimulus in the monkey's visual field and found strong evidence to the contrary. We compared tuning for the circle's location during the 800 ms before the circle started moving (called the early period) and during the last 800 ms of the delay period, immediately prior to the trigger signal (the late period). (Figures S5 and S6 show activity during a slightly different early period than measured here, but they illustrate the same basic result.) Despite the fact that the sensory inputs were identical in screen-centered, allocentric, retinocentric, fixation-centered, head-centered, and body-centered coordinates, the activity of PFdl neurons and their degree of spatial tuning differed in these two task periods. This result rules out a purely sensory response. For the entire PFdl sample, the late tuning index (1.29 ± 0.03) significantly exceeded the early one (1.16 ± 0.02; p < 0.001; Wilcoxon matched-pairs test). This measure is devoid of any bias caused by a cell's tuning properties in one task period or the other, but it includes the contribution of the spatially untuned cells. When we restricted the comparison to neurons that had any type of significant spatial tuning, in either the early or late periods, the late tuning index (1.76 ± 0.07) continued to exceed the early one (1.42 ± 0.05) significantly ( p < 0.001). Most important, we obtained similar results for neurons with significant tuning to the circle's location, which characterizes attention and hybrid cells (1.83 ± 0.08 late versus 1.46 ± 0.05 early; p < 0.001). Table 3 and Figure S4 present this analysis for all cell classes, alone, and in various combinations. Note that these indexes do not reflect a generalized increase in firing rate: They were normalized to remove the effects of firing rate per se. The section entitled Population Analysis presents a confirmatory result in terms of activity levels. Further confirming this result on a cell-by-cell basis, significant spatial tuning to the circle's location occurred more frequently during the late delay period (256 attention and hybrid cells) than during the early one (194 cells, of which 41 lost their spatial tuning in the late period). Thus, the representation of the circle's location in PFdl grew stronger around the time of the trigger signal, when it was important for the monkeys to attend to the circle. These findings rule out the mere presence of the circle in something akin to a visual receptive field as a complete account of the tuning of attention and hybrid cells. Histological Analysis Figures 4 and 5 show the locations of the cells in each class: Figure 4 as a function of electrode-penetration sites for both monkeys and Figure 5 as section reconstructions for monkey 2. The attention cells were concentrated more ventrolaterally than either the memory or the hybrid cells. Neurons located ventrolateral to the fundus of the principal sulcus ( n = 551) were predominantly attention cells (28% to 2% memory and 5% hybrid cells, with 65% lacking spatial tuning, both monkeys combined). Neurons dorsomedial to the fundus ( n = 412) fell into the three cell classes approximately equally (8% attention, 9% memory, and 10% hybrid cells, with 73% lacking spatial tuning). These regional differences within PFdl were highly significant for each monkey (p < 0.0001, χ 2 test). Cells with significant memory signals (memory and hybrid cells, combined) composed 70% of the spatially tuned population in dorsomedial PFdl, but only 20% in ventrolateral PFdl. Figure 4 Surface Projections Showing the Location of Neurons in Each Class All hemispheres are displayed so that rostral is to the left and dorsomedial is up. Reconstructed surface projections of the left hemispheres of monkey 1 (A) and monkey 2 (B). (C) Surface projection of the (inverted) right hemisphere of monkey 1. (D) A lateral view of the hemisphere shown in (C), with the region included in (C) approximated by the dashed box. The dotted ellipse encloses the cells deemed to lie inside the PFdl by histological analysis, but does not correspond to the cytoarchitectonic boundaries of area 46. Figure 5 Section Reconstructions for Monkey 2 (A–G) Coronal sections taken at the planes indicated in the surface drawing (H). Dashed lines mark the borders between PFdl (area 46) and area 12. Solid lines show the tracks of the marking pins (irregular outlines in sections [B] and [C]) and the estimated location of electrode penetrations. Colored hash marks show the estimated depth of neurons in each class, using the same color code as in Figures 3 and 4 . Longer hash marks indicate simultaneous recordings of more than one neuron of the same class. (H) Lateral view of left PF depicting surface projections of spatially tuned neurons. Black circles show the locations of pin holes used for localization, and gray squares show their predicted locations. Based on a cytoarchitectonic analysis conducted on two of the three hemispheres, all of the cells situated ventrolateral to the fundus of the principal sulcus were located within area 46 and none were located in area 12. The area 46/12 architectonic boundary was first described by Walker (1940) and was subsequently confirmed with different methods ( Preuss and Goldman-Rakic 1991 ). This boundary could be discerned in both monkeys as a distinct thinning of the internal granular layer in area 12 compared to area 46 and a more substantial departure in that area from the classic, homotypical appearance typical of area 46. The reconstructed location of recording sites showed that the small group of cells located caudomedially in both monkeys (see Figure 4 B and 4 C) was located in the postarcuate cortex (area 6) and in area 8, as indicated by the agranular and dysgranular cytoarchitecture of these two regions, respectively. This small group of cells was eliminated from the present analysis. Population Analysis Figure 6 displays the degree of spatial tuning for the different cell classes in the form of population histograms. The analysis of attention tuning ( Figure 6 A and 6 B) used the 800 ms immediately preceding the trigger signal to measure mean firing rates for different attended locations. We excluded control trials from this analysis. These rates were then ranked from the largest (i.e., the preferred attended location) to the smallest (the least preferred location). For each neuron, the preferred location chosen by this analysis was designated as preferred for all task periods displayed in the population histograms. (Similar results were obtained when the ranking was done for each individual task period.) The left side of the figure shows the mean attention signal for both attention ( Figure 6 A) and hybrid ( Figure 6 B) cells. After a transient response to the appearance of the circle (at a latency of approximately 100 ms), neuronal activity in both of these cell classes remained elevated when the circle stopped at the preferred location (blue curve) and became slightly suppressed when it was at the least preferred location (black curve). Figure 6 Attention and Memory Signals in Population Histograms (A) and (B) Representation of an attention signal by attention cells (A) and hybrid cells (B). (C) and (D) Representation of a memory signal by memory cells (C) and hybrid cells (D). In each panel, activity is shown centered on the appearance of the circle (left vertical line), on the time that the circle stopped moving (middle vertical line), and on the trigger signal (right vertical line). Attention and memory signals are reflected in the degree of separation in the average population histograms for different ranks. In (A) and (B), the data for the period immediately prior to the end of the circle's movement have been eliminated because the circle came from different initial locations. The right side of Figure 6 shows the mean memory signal for memory ( Figure 6 C) and hybrid ( Figure 6 D) cells. These population histograms were calculated on the basis of preferred remembered locations, ranked according to the pretrigger modulation. This location was then designated “preferred” for all task periods displayed in the plots. For memory cells ( Figure 6 C), the population averages were almost identical when the circle remained stationary at its initial location and that location did not yet need to be remembered. That is, on average it did not matter noticeably whether the circle initially appeared at a cell's preferred location or at its least preferred location ( Figure 6 C, red versus black curves). This finding is somewhat surprising because prior studies suggested that PFdl's memory cells had activity that began shortly after stimulus onset and continued throughout the delay period. In our memory cells, spatial tuning did not develop to any appreciable extent until after the circle began revolving around the central fixation point. This result shows that tuning to the remembered location developed during the trial and was not a simple replica of the tuning pattern during the initial presentation of the circle. Hybrid cells ( Figure 6 D) exhibited a weak spatial signal following the appearance of the circle consistent with their memory tuning prior to the trigger. Note that after the circle stopped moving, memory cells showed less of a difference between preferred and least preferred locations than did attention cells ( Figure 6 C versus 6A). This finding supports the results presented in Tables 2 and 3 and Figure 3 D, which show a predominance of nonmemory signals (see also Figure S4 ). Population representations of the attended and remembered locations were further analyzed using a neuron-dropping analysis. Neuron-dropping curves express the strength of spatial tuning as the ability to estimate a spatial variable from the activity of a neuronal ensemble, as a function of ensemble size. We randomly selected an ensemble from the population of recorded PFdl neurons and used a single trial of activity from each cell to estimate both the attended and remembered locations. The findings of the neuron-dropping analysis agree with those from the analysis of single-cell activity and the population histograms and thus provide independent support. However, neuron-dropping analysis offers several advantages over the population histograms, in addition to providing confirmation of those results. In neuron-dropping, the estimation of either an attended or remembered location does not depend on any assumptions about the nature of the spatial tuning curve or the relative importance of very active cells versus those showing less activity. It does not ascribe any special significance to increases in activity relative to baseline (excitation) versus decreases (inhibition) or to the most preferred and least preferred locations. Each cell's activity contributes to the population estimation for all locations regardless of the direction of its modulation relative to baseline and whether that modulation significantly differs from baseline levels. Furthermore, the computation makes no assumption about any relationship between tuning for attended locations and remembered ones. This analysis also has the advantage that its results are expressed as a percentage of correct estimations by the neuronal ensemble, thereby facilitating comparison with the monkeys' performance, which in this experiment always exceeded 75% correct and sometimes approached 100% ( Table 1 ). Figure 7 shows the neuron-dropping curves for each cell class (A–C) and all spatially tuned neurons combined (D) in monkey 1. Neuron-dropping curves for monkey 2 showed similar results, and Figure S7 presents the data for both monkeys combined. As expected, the neuron-dropping curves computed for attention cells yielded much better estimations of the attended location than the remembered one (see Figure 7 A, blue versus red curves). Note, however, that the attention cells also provided a better-than-chance estimation of the remembered location. This result reflects the fact that many cells with significant tuning for the attended location also showed some tuning for the remembered location (see blue data points in Figure 3 D with I Rem > 1.0). Figure 7 A also confirms the comparison of activity early versus late in the trial (blue versus gray curves), providing further evidence against a purely sensory account of this subpopulation's activity. Also as expected, memory cells yielded a better estimation of the remembered location than the attended one ( Figure 7 B, red versus blue curves), but these cells, too, yielded a fairly reliable estimation of the other spatial variable. Neuron-dropping curves for hybrid neurons showed comparable estimations for both locations ( Figure 7 C). When all spatially tuned neurons were combined ( Figure 7 D; see also Figure S7 D), the resultant neuron-dropping curves showed that PFdl activity was a much more reliable estimator of the attended location than the remembered one. Figure 7 Neuron-Dropping Curves for Different Subpopulations of PFdl Neurons in Monkey 1 Each curve represents the percentage of correct single-trial estimations of location as a function of the number of neurons in the assembled populations. The curves show predictions of the attended locations (blue lines) or remembered locations (red lines) during the 800 ms immediately preceding the trigger signal, after the circle had stopped revolving around the central fixation point. Also shown is the estimation for the 800-ms period immediately preceding the onset of the circle's movement (gray lines). The dotted line indicates the chance level of estimation, 25% correct. Neuron-dropping curves are shown for neurons tuned to the attended location (A), the remembered location (B), both locations (C), and all spatially tuned neurons (D). (E) and (F) Dynamic changes in estimations of the attended (blue) and remembered (red) locations for 20 spatially tuned neurons (marked by the dashed gray line and arrows), using a 200-ms sliding window. Dashed and solid lines in (E) and (F) are shown for consistency with Figure 8 . Note that the estimations in (D) are higher than in (E) and (F) because the former is based on an 800-ms interval, and the latter are based on only a 200-ms interval. The same analysis was applied to the ventromedial and dorsolateral regions within the PFdl, described in the section entitled Histological Analysis, above (not shown). The ventrolateral subpopulation of PFdl neurons (see Figure 4 A– 4 C) overwhelmingly represented the attended location. The dorsomedial subpopulation represented both locations comparably, with estimation of the attended location being slightly better in one monkey and estimation of the remembered location being slightly better in the other. Of the two subpopulations, the dorsomedial neurons showed a more reliable estimation of the remembered location. We also used a neuron-dropping analysis to examine the ensemble's properties during response selection and execution. Figure 7 E and 7 F show these time-dependent neural-estimation curves for monkey 1; Figure 8 does so for both monkeys combined. Note from Figure 7 D– 7 F that the time-estimation curves come from a random sample of neurons, much smaller than the sampled population, to avoid the effects of signal saturation. The estimations at each time point reflect activity averaged over the previous 200 ms. Prior to the trigger signal, the estimation of the attended location (blue curves in Figures 7 E, 7 F, 8 D and 8 E) was superior to that of the remembered location (red curves) for all spatially tuned neurons, as well as for attention cells ( Figure 8 A). This finding is consistent with the greater number and stronger spatial tuning of attention than memory cells. Figure 8 Time-Dependent Changes in Estimating the Attended Location and Remembered Location, for Both Monkeys Combined Solid lines, trials in which the monkeys made a saccade to the attended location; dashed lines, trials in which the monkey made a saccade to the remembered location. Blue lines, estimation of the attended location; red lines, estimation of the remembered location. All records are centered on the onset of the trigger signal (using data for the 200 ms prior to that time). Vertical lines at t > 0 show the average saccade latency on Att-trials (solid) and Rem-trials (dashed). The thick bar at the bottom of the plots shows the approximate onset of the peripheral fixation spot, which the monkeys continued fixating beyond the limit of the plot. Estimations for each monkey were calculated using the same methods as for Figures 7 E and 7 F, except that the ensemble size for monkey 2 was 60 neurons. This number of neurons was chosen to avoid ceiling effects (i.e., 100% correct). The plotted curves show the average for the two monkeys. Location estimations for attention (A), memory (B), hybrid cells (C), and all spatially tuned neurons on Att-trials (D) and Rem-trials (E). Above the plots are schematic depictions of an example trial, of the type illustrated in Figure 1 A. The red “R” marks the remembered location. In both D and E, prior to the trigger signal (left schematic), the monkeys fixated (dashed lines) centrally and covertly attended to the circle (yellow spot at the attended location). During this period, estimation of the attended location exceeded that of the remembered location. Following the saccade, the monkeys fixated a peripheral light spot (right schematic) and attended to this target to detect when it dimmed. On Att-trials (D), the monkeys' gaze shifted to the attended location, and the ensemble's estimation of this now overtly attended location improved (solid blue curve), while the representation of the now irrelevant, remembered location gradually decayed (solid red curve). On Rem-trials (E), the monkeys' gaze (dashed lines) and focus of attention (yellow spot) shifted to the (previously) remembered location. The estimation of this location consequently improved (red dashed curve), while the estimation of the previously attended (and now irrelevant) location gradually decayed. Abbreviations: Att, attended; Rem, remembered. On Att-trials, the estimation of the attended location (solid blue curves in Figures 7 E and 8 A–D) improved following the dimming of the circle and remained elevated during the saccade to that location. This improvement continued for the initial 200 ms of fixation there. Then the signal decreased. Note that the monkey maintained fixation at the target location for at least 1.0 s after the saccade. In contrast, the estimation of the remembered location on Att-trials (solid red curves) gradually decreased following the trigger signal. The fading of this representation most likely reflected the fact that the remembered location was no longer behaviorally relevant. On Rem-trials (dashed curves in Figures 7 F, 8 A– 8 C, and 8 E), the circle's brightening instructed a saccade to the remembered location (marked by the red “R” in Figure 8 E). We expected that redirecting attention toward the saccade target (yellow spot in Figure 8 E, right) would degrade the neuronal representation of the formerly attended location and improve the representation of the formerly remembered—but eventually fixated—one. The estimation of the attended location initially improved on Rem-trials following the trigger signal there (blue dashed curves in Figures 7 F, 8 A– 8 C, and 8 E). However, in accord with our expectation, that estimate decreased dramatically in accuracy after saccade onset, as the attended location became behaviorally irrelevant. In contrast, the estimation of the formerly remembered (and soon to be fixated) location (red dashed curves) improved sharply ( Figure 8 E), especially in attention cells ( Figure 8 A). Thus, PFdl neurons became more reliable encoders of that location. Given that these averages “look back” 200 ms, this development must have preceded the saccade. On both Att-trials and Rem-trials, the neuronal ensemble remained a reliable indicator of the saccade target relatively long after the target had been acquired (see solid blue and dashed red curves in Figures 7 E, 7 F, and 8). This signal might encode the fixated location, which could be important for monitoring performance, as suggested for nearby areas of frontal cortex ( Stuphorn et al. 2000 ; Ito et al. 2003 ). Alternatively, the saccade target may have been represented because the monkeys attended to the fixation spot at this location, so that when it dimmed they could quickly release the button to produce their reward (see Materials and Methods , below, for a description of that aspect of the task). Discussion In tasks involving short-term memory requirements, delay-period activity in PFdl has consistently been interpreted in terms of the maintenance-memory theory of PF function (e.g., Funahashi et al. 1989 ; Romo et al. 1999 ; Constantinidis et al. 2001 ), despite the existence of viable alternatives. However, our results show that much of PFdl's delay-period activity in such tasks reflects nonmemory functions. Accordingly, the maintenance-memory theory of PF function ( Goldman-Rakic 1987 , 1990 ), taken to its extreme, fails to account for PFdl's delay-period activity. Indeed, we found that, compared to the remembered location, the attended location was more frequently and more robustly encoded at both the neuronal and population levels. The present results thus support extensive neuropsychological ( Rueckert and Grafman 1996 ; Stuss et al. 1999 ; Koski and Petrides 2001 , 2002 ) and neuroimaging ( Corbetta et al. 1993 ; Gitelman et al. 1999 ; Kastner et al. 1999 ; Rosen et al. 1999 ; Cabeza and Nyberg 2000 ; Hopfinger et al. 2000 , 2001 ; Vandenberghe et al. 2000 ; Astafiev et al. 2003 ; Small et al. 2003 ; Thiel et al. 2004 ; Woldorff et al. 2004 ) research that points to a much more general role for PF than encompassed by the maintenance-memory theory, including the top-down control of selective attention. Interpretational Issues and Limitations The present experiment is the first neurophysiological study to achieve a degree of independent control over both spatial attention and spatial memory, so a detailed consideration of both its advantages and limitations is in order. A complete dissociation of these two spatial variables is probably impossible, but we achieved this goal to a considerable degree. Our experimental design, however, has several limitations and raises a number of questions. For example, is what we call attention really attention? We have elaborated on our usage of the term attention in the Results section. Although we did not quantify the degree of attention, it seems to us a reasonable assumption that the monkeys attended to the circle, given that its brightening or dimming was subtle, brief, and crucial to their correct performance. Moreover, the reaction-time data are consistent with the idea that the monkeys attended to the circle in the period immediately prior to the trigger signal. The remaining interpretational questions to be addressed, then, are: Do monkeys devote any attentional resources to what we call the remembered location? Do they “remember,” in some sense, what we call the attended location? Does the activity we interpret in terms of attention or memory reflect motor factors? And, given that the monkeys could anticipate and predict rewards, do the signals reflect these processes? We address each of these four questions, in turn, in the remainder of this section. First, although we contend that the monkeys must have devoted substantial attentional resources to the location of trigger signal, this does not necessarily rule out additional covert allocations of attention to the remembered location. However, there was no stimulus or expected signal at the remembered location to warrant the allocation of attentional resources there. In addition, the demands of fixating the central location (overt attention), while attending covertly to a stimulus located in peripheral visual space, make it unlikely that attention was further divided ( Hunt and Kingstone 2003 ; Muller et al. 2003 ). Accordingly, although we cannot completely rule out the possibility that the monkeys attended to the remembered location during the delay period, it seems implausible that they did so. If one adopts the view that they did, then some or all of the neurons we class as memory cells might instead have activity better interpreted as reflecting some aspect of highly divided attention. Second, the monkeys were required to remember the place where the circle first appeared on each trial, and their performance shows that they did so. Did they also “remember” the attended location? There is ample precedent for skepticism about the proposition that monkeys are not remembering some location. However, there is no basis for assuming a “memory” of a currently visible stimulus. It seems especially unlikely that the monkeys “remembered” the attended location in the context of the requirement that they centrally fixate while attending somewhere and remembering somewhere else. Third, we cannot rule out the participation of neurons we class as attention or memory cells in a variety of processes involved in preparing or planning the movement or selecting the response target. Prior to the trigger signal, the monkeys may have prepared to make a movement to the remembered location, to the attended location, to both, or to neither. Cisek and Kalaska (2002) have shown that some neurons in the premotor cortex encode a possible movement target before a particular one has been specified, but their experiment has yet to be done for PFdl neurons. In view of prior evidence arguing against interpreting much of PFdl's delay-period activity in terms of motor signals ( Funahashi et al. 1989 , 1993b ; di Pellegrino and Wise 1993b ; Asaad et al. 1998 ; Romo et al. 1999 ; Constantinidis et al. 2001 ) and the absence of a contemporary “motor theory” of PF function, the present experiment was not designed to address this issue. Future work along these lines, perhaps combining the design of di Pellegrino and Wise (1993b) with the present one, might be indicated by the present results. We believe, however, that a simple “motor” explanation for most of PFdl's delay-period activity is an unlikely outcome of such studies. A “motor” interpretation probably does, however, account for a small proportion of PFdl's delay-period activity, consistent with the results of Funahashi et al. (1993b) . On certain assumptions about a default motor plan, such neurons could have the tuning properties of the hybrid cell illustrated in Figure 3 C, part a. It is important to emphasize, however, that the present experiment tested whether the maintenance-memory theory could account for all delay-period activity in PFdl. It cannot. We view this result as supporting an important role for PF in the top-down control of attention. If one takes a motor theory of PF function more seriously than most expert opinion currently does, then it is possible to interpret the present result as indicating a role in context-dependent response or goal selection or in terms of the preparation of movements to remembered targets versus current stimuli. Neither interpretation is consistent with an interpretation of PFdl's delay-period activity entirely in terms of a maintenance-memory function. Fourth, we need to consider the possibility that the neural signals we observed reflect the prediction or anticipation of reward. Maunsell (2004) has recently pointed out that neural signals interpreted as arising from attention could instead reflect reward anticipation or prediction (and vice versa). In the present study, however, reward-related information processing could not have accounted for the properties of attention cells because, until the trigger signal, one alternative place (the remembered location) was associated with reward to the same degree as the attended location. Enhancement Effects The general term attention has been used to cover many disparate concepts, including the effects of attention on sensory processing and the mechanisms that mediate those influences. We emphasize that the present finding differs from previous ones describing effects of attention on phasic, sensory-like responses. Often called the enhancement effect, the finding that sensory responses are larger when a stimulus or location is more attended was first described for the superior colliculus ( Wurtz and Goldberg 1972 ) and has been repeatedly demonstrated for many cortical areas, including PFdl ( Mikami et al. 1982 ; Boch and Goldberg 1989 ; di Pellegrino and Wise 1993a ; Rainer et al. 1998 ; DeSouza and Everling 2004 ). In some instances, and especially in frontal cortex, the enhancement effect depends on the attended location being the target of a movement ( Goldberg and Bushnell 1981 ), but in other cases it does not ( Bushnell et al. 1981 ). It has often been suggested that the source of attention effects, including the enhancement effect, match enhancement, and related phenomena, depends on signals emanating from PF ( Miller et al. 1996 ; Kastner et al. 1999 ; Reynolds et al. 1999 ) or from the frontal eye field ( Thompson et al. 1997 ; Moore and Fallah 2004 ). The present results are consistent with this idea. They cannot, however, be considered as yet another example of the enhancement effect, which involves attention-dependent augmentation of a phasic sensory response. Neurons Encoding Both Attended and Remembered Locations Most neurons did not encode an attended or remembered location exclusively; rather, they exhibited varying degrees of tuning for both variables. The neuron-dropping curves (see Figure 7 ) show that attention cells were able to make limited, but above-chance, estimations of the remembered location and vice versa. As can be seen from the spatial tuning indexes in Figure 3 D, few individual neurons were pure attention or memory encoders (data points along the axes). Thus, the population of spatially tuned cells can be viewed as a continuum with attention and memory cells at the extremes, and hybrid cells in between. Interestingly, the neuron-dropping curves for the hybrid cells (see Figures 7 C and 8 C) showed effective estimation of both the attended and remembered locations. Hybrid neurons with dissimilar preferences for the two locations facilitated such estimations. For instance, the neuron shown in Figure 3 C, part b had a low firing rate when the monkey attended to the 180° location and a high firing rate when it remembered that place. Hybrid cells with dissimilar preferences can resolve the ambiguity inherent in cell activity like that illustrated in Figure 3 C, part a, which cannot distinguish between attended and remembered locations. Previous Neurophysiological Studies Previous neurophysiological studies of PFdl's delay-period activity have been interpreted in terms of the maintenance-memory theory. However, the lack of control over spatial attention in these studies raises questions about these interpretations. Constantinidis et al. (2001) , for example, trained monkeys to make delayed saccades toward the location of the brighter of two visual stimuli that briefly flashed on the video screen. They reported that the activity of PFdl neurons reflected the brightness of the stimuli. Although these authors interpreted their findings as demonstrating a purely sensory-mnemonic function for PFdl neurons, brighter stimuli, being more salient, are well known to attract attention to their location. Similar problems affect the interpretation of data from the “antisaccade” task ( Funahashi et al. 1993b ). In their antisaccade task, Funahashi et al. trained a monkey to respond to a stimulus to the left of a fixation point by making a saccade to the right and vice versa. They interpreted their data as demonstrating a function for PFdl in spatial memory because the largest number of neurons reflected the stimulus location rather than the movement target. They showed that during the delay period, when nothing was present on the screen, some neurons reflected where the stimulus had occurred, and these were interpreted as memory cells. Note, however, that where ever the stimulus appeared, whether in antisaccade or prosaccade trials, it served as an attention attractor. If the response to that signal persisted, then interpreting it exclusively as a sensory memory trace would be problematic. Many studies suggest that, for neurons in PF, the history of what has happened or the context in which it happens often affects neuronal activity in an important and persistent way ( Rainer et al. 1998 ; Asaad et al. 2000 ; Wallis and Miller 2003 ), sometimes regardless of relevancy ( Chen et al. 2001 ). Such persistent signals can be viewed as components of working memory in a general sense, but not in the narrow sense implied by the concept of maintenance memory. Neuroimaging and Neuropsychological Results from Humans Based on the idea that the principal or exclusive function of PFdl is to support maintenance memory ( Goldman-Rakic 1987 ), many neuroimaging papers on PF, including PFdl, have been interpreted as supporting this theory of PF function (see, for example, Courtney et al. 1996 , 1997 , 1998 ; Druzgal and D'Esposito 2003 ; Inoue et al. 2004 ). This idea has been defended ( Goldman-Rakic 2000 ), but a number of alternatives have been suggested. For example, several neuroimaging findings support a role for PF in the control of attention, and brain lesion studies also show attentional deficits after damage to various parts of PF ( Corbetta et al. 1993 ; Rueckert and Grafman 1996 ; Gitelman et al. 1999 ; Kastner et al. 1999 ; Rosen et al. 1999 ; Stuss et al. 1999 ; Cabeza and Nyberg 2000 ; Hopfinger et al. 2000 , 2001 ; Vandenberghe et al. 2000 ; Koski and Petrides 2001 , 2002 ; Astafiev et al. 2003 ; Small et al. 2003 ; Thiel et al. 2004 ; Woldorff et al. 2004 ; see also a recent review by Wood et al. 2003 ). In general, top-down attention has been assumed to result from signals emanating from the frontal cortex and biasing more posterior areas to favor some channels of information over others, and some neuroimaging papers have supported this idea ( Chawla et al. 1999 ; Kastner et al. 1999 ; Corbetta and Shulman 2002 ; Nakahara et al. 2002 ; Pessoa et al. 2003 ). In addition, a role in attentional selection and the related concepts of attention to action and attention to intention have been stressed as an alternative to the maintenance-memory theory of PF function ( Rowe et al. 2000 ; Rowe and Passingham 2001 ; Lau et al. 2004 ). Similarly, monitoring the items in short-term memory has been put forward as a principal function of PFdl, and this also is primarily an attentional function ( Owen et al. 1996 ; Petrides et al. 2002 ). Along these lines, a recent study by Nobre et al. (2004) indicated that PF plays a role in directing attention to locations within mental representations. Neuropsychological Results from Monkeys Previous research on monkeys has also suggested a role for PF (or nearby parts of the frontal lobe) in the orientation of spatial attention. Welch and Stuteville (1958) produced trimodal (auditory, visual, and tactile) neglect-like effects following ablations in the depths of the arcuate sulcus, including what was likely part of PF (although not PFdl). Rizzolatti et al. (1983) reported neglect for space beyond a monkey's reach after lesions targeting area 8. However, for at least one of the two monkeys they studied, the lesion may have included the area studied here. Deuel and Farrar (1993) also produced neglect-like symptoms by making cortical lesions that included much of the same region, and roughly similar observations have been interpreted as motor neglect ( Heilman et al. 1995 ). PF lesions also caused attention-like deficits in a conditional motor learning task (M.F.S. Rushworth et al., personal communication). In the context of the present results, the finding that inactivation of parts of PFdl ( Funahashi et al. 1993a ) produced what were termed “mnemonic scotomas” deserves reconsideration. In that experiment, a transient cue served as the target of a saccade after a delay period. Following local inactivations within PFdl, the monkeys in that study continued to make most of their responses to sites near the cue's remembered location, even with 3-s and 6-s delays after the disappearance of the cue (see their Figures 5, 9, and 13). The monkeys made the vast majority of their responses in the correct direction, but a few saccades fell outside the target zone. This inaccuracy contributed to significantly increased variance in the endpoints of the saccades, and Funahashi et al. (1993a) concluded on this basis that the monkeys were unable to remember the cue's location. We suggest, as an alternative explanation of their results, that their monkeys had a deficit in detecting the stimulus at the cued location, directing attention there, or maintaining their attention at the cued location. Thus, the results interpreted as “mnemonic scotomas” might be better understood as a localized neglect-like phenomenon or some combination of attention and memory deficits. This suggestion finds support in the results of a recent study in humans with PF lesions. Hornak et al. (2004) reported a failure of such patients to pay attention to information on a screen, and this problem accounted for their behavioral deficits. Therefore, the results of Funahashi et al. (1993a) provide little support for either the maintenance-memory theory of PF function or the interpretation of its delay-period activity in terms of that theory. The present results agree better with those of Rushworth et al. (1997) , who found that monkeys could remember nonspatial stimuli across relatively long delay periods after bilateral removal of the part of PF theorized to maintain such memories. The present results also agree with Petrides (2000) , who found that PFdl lesions do not affect the short-term memory for objects (as measured by a susceptibility to increasing delay periods), but do cause impairments in the ability to monitor which items have been selected from a group (as measured by a susceptibility to increasing group size). Conclusions The present study reexamined the interpretation of PFdl's delay-period activity in terms of the maintenance-memory theory. We found that other factors are more important than mnemonic ones. The present results do not argue against a short-term memory function for PF, as one among many contributions to behavior. Nor should they lead to the dismissal of interpretations of some delay-period activity in PF, or some neuroimaging signals from that region, in terms of short-term memory. However, spatial memory signals occur less frequently in PFdl than the maintenance-memory theory predicts. Our data thus accord better with neuroimaging and neuropsychological studies indicating that PF plays a major role in attentional selection, including the monitoring of information and actions ( Owen et al. 1996 ; Rowe et al. 2000 ; Rowe and Passingham 2001 ; Petrides et al. 2002 ; Manly et al. 2003 ; Lau et al. 2004 ). How do our findings mesh with the fact that damage to PF appears to produce deficits in short-term memory, as Jacobsen (1935 , 1936 ) first showed nearly 70 years ago? One possibility is that lesion studies speak more to the inability of other areas to compensate for the loss of PF than to the priority of functions within that region. Another is that an attentional deficit would likely have an important effect on the performance of tasks typically used to assess short-term memory in monkeys, such as matching-to-sample or delayed-response tasks, especially if monkeys use selective attention as a strategy for solving the problems posed by such tasks (see di Pellegrino and Wise 1993b ; Awh and Jonides 2001 ). Although attention could account for many findings about PF, we do not aim to replace one monolithic theory of PF function—the maintenance-memory theory—with an equally monolithic “attention theory.” Delay-period activity appears to reflect the learning and implementation of behavior-guiding rules ( Wise et al. 1996b ; White and Wise 1999 ; Wallis et al. 2001 , Wallis and Miller 2003 ), categorization of events and stimuli ( Freedman et al. 2001 , 2003 ), prediction of forthcoming events ( Rainer et al. 1999 ), task selection ( Hoshi et al. 1998 ; Asaad et al. 2000 ), and adaptive actions within structured-event sequences ( Barone and Joseph 1989 ; Quintana and Fuster 1999 ; Ninokura et al. 2003 , 2004 ; Hoshi and Tanji 2004 ), among other cognitive functions. According to one view, PF functions in general intelligence for the solution of any and all difficult cognitive problems ( Duncan and Owen 2000 ). Gaffan (2002) has likewise argued that PF resembles a global workspace, in the sense used by Baars et al. (2003) , implying a lack of domain selectivity. The present result, by showing that PFdl's delay-period activity lacks an account solely in terms of maintenance memory, supports these ideas to some extent. However, the finding of regional specializations among different parts of the PFdl (see Figure 4 ), in accord with similar findings ( Ninokura et al. 2003 , 2004 ; Hoshi and Tanji 2004 ), suggests that various parts of PF contribute to this global workspace differently, each by making some selective contribution to PF's overall function. Taken together, these observations suggest that delay-period activity in PF reflects functions extending far beyond maintenance memory to include all of the behaviors important to the life of primates. Materials and Methods Behavioral task, apparatus, and single-unit recordings We trained two rhesus monkeys (Macaca mulatta) to perform the task. Each monkey sat in a primate chair in front of a computer monitor placed 57 cm from the monkey's eyes. We recorded eye position with an infrared oculometer and sampled at 250 Hz. The monkeys pressed a waist-high button with their right hand to start each trial and did not release the button until the end of the trial. Once the monkeys pressed the button, a 0.2° fixation point appeared at the center of the screen. After they had fixated this stimulus for 1.0–1.5 s, a 2° solid, gray circle appeared 8° from the center of the screen in one of four places. Figure 1 A, part a illustrates the right (0°) location. After another 1.0–1.5 s, the circle revolved from this initial location to one of four final places ( Figure 1 A, part b) at 90°/s along a circular trajectory centered on the fixation point. For monkey 1, the circle revolved 90° or 180° either clockwise or counterclockwise. For monkey 2, the circle revolved 90°, 180°, or 270° either clockwise or counterclockwise. After the circle stopped, a 1.0 to 2.5-s delay period ensued. Then a trigger signal occurred, which provided an instruction as to the saccade target, as well as a “go” cue for the saccade. The trigger signal consisted of a 150-ms-long change in the circle's brightness ( Figure 1 A, part c), followed by its disappearance ( Figure 1 A, part d). If the circle dimmed, the saccade had to be directed to the circle's final (and current) location on that trial; if the circle brightened, the saccade had to be directed to the circle's initial location on that trial. After the monkeys started a saccade, the central fixation spot disappeared. If the monkeys made a saccade to the correct location, a new 0.2° fixation spot appeared there, and the monkeys had to fixate this spot for 1.0–1.5 s, after which it dimmed. The monkeys could then release the button to produce a fruit juice reward. If the monkeys broke fixation prior to the trigger signal, made an incorrect saccade, or released the button prematurely, the trial was cancelled, and the monkeys could begin a new trial. In control trials, the circle either did not move (both monkeys) or returned to its initial location (360° movement, either clockwise or counterclockwise, monkey 2 only), and the monkeys had to make a saccade to the location of the circle whether it dimmed or brightened. The initial and final locations of the circle and whether it brightened or dimmed were selected pseudorandomly, as was the duration of the delay period and the direction in which the circle revolved around the central fixation point. The monkeys had to complete one correct trial of each type (32 in all, including control trials) before repeating a trial type. After the monkeys learned the task, we implanted recording chambers over the left (monkeys 1 and 2) and right (monkey 1) PFdl. For monkey 1, we used a single-electrode microdrive to obtain single-neuron activity records; for monkey 2, we used a microdrive that independently moved up to seven electrodes. During recordings in monkey 1, we intentionally biased the selection of task-related neurons toward those with delay-period activity. In monkey 2, we recorded the activity of all isolated neurons, regardless of whether they were task related. For histological reconstruction of recording sites, we examined Nissl-stained sections of 40 μm thickness from the right hemisphere in monkey 1 and the left hemisphere in monkey 2. Quantification of tuning We represented firing rate data in a 4 × 4 matrix, F ij , with rows (i) corresponding to the remembered location and columns (j) to the attended location ( Figures 2 , 3 A– 3 C, S1–S3, S5, and S6). We assessed tuning for the remembered locations by comparing the variability of firing rate between trials in different rows with the variability of firing rate between trials from the same row. To avoid the influence of across-column modulations (i.e., an attention effect), both between-row and within-row variabilities were calculated only for matrix elements from the same column, one column at a time, and then we averaged these results. This procedure amounts to comparing different remembered locations, while holding the attended location fixed. To quantify the strength of tuning for the remembered location, we computed a ratio of between-row variability and within-trial type variability: (1) where l 1 and l 2 index individual trials, i, j, and k are matrix indexes that take on the values of 0°, 90°, 180°, and 270°, F ij (l) is the firing rate on the l th trial for which position i was the remembered location and position j was the attended location, and N 1 and N 2 are total number of elements in the respective sums. Control trials were excluded from the calculation by not considering the diagonal elements of F ij (i = j, j = k). We evaluated tuning to the attended location similarly by comparing across-column variability with within-column variability, one row at a time. The strength of representation of the attended location, was quantified as: (2) We used two task periods to compute the single trial firing rates F ij (l). To classify neurons into those representing remembered versus attended location, we used an 800-ms period preceding the trigger signal. We also evaluated spatial tuning (I) during the final 800-ms period before the circle started to move. (Figures S5 A and S6 A illustrate this “early” period slightly differently, averaging activity in an interval from 200 ms to 1,000 ms after the appearance of the circle.) The I Rem or I Att ratios approximated unity for untuned neurons and increased with tuning strength. To measure statistical significance, we used a randomization test. Trials were randomly shuffled among different remembered or attended locations, and the indexes were recomputed. This procedure was repeated 1,000 times to yield a distribution of index values from which we computed a probability p. We chose a statistical significance level of p < 0.01 to classify neurons as tuned for either I Rem , I Att , both, or neither (untuned). Population histograms We computed the population histograms of Figure 6 by first determining each neuron's preferred location, using firing rates during the 800 ms preceding the trigger signal. Then we ranked the trials as belonging to the preferred location, the next most preferred, the third most preferred, and the least preferred location. This ranking was then applied to the other task periods. Next we calculated peri-event time histograms (PETHs) for each rank, separately for each neuron. Averages of these single-neuron PETHs yielded the population histograms. To avoid biasing average histograms by statistical noise in the ranks, we used one half of the trials to compute the ranks and the other half to compute the histograms. If the spatial preference of a neuron merely reflected noise, this procedure tended to nullify the influence of the neuron on the population average. We ranked attended and remembered locations in separate computations. Neuron-dropping curves Neuron-dropping curves ( Figures 7 A– 7 D and S7) estimated how well ensembles of PF neurons represented the remembered and attended locations of the circle ( Wessberg et al. 2000 ). We excluded control trials, in which the circle either did not move or moved 360°, from this analysis. The method measured the probability that the attended and remembered locations could be correctly estimated using a single trial of activity from a neuronal ensemble as a function of its size. The calculation started with a random selection of n neurons from a population. Then, for a given condition (e.g., a remembered location i of 0° and an attended location j of 90°), we selected one trial of that condition randomly from each neuron (test trials). All the other trials for that neuron contributed to a look-up table of firing rates. This look-up table consisted of a matrix of average firing rates <F ij > for remembered locations, i, and attended locations, j. The differences between firing rates in the look-up table and the rate on the selected trial were rank ordered, with a smaller rank signifying a closer match. We then summed the ranks r ij across individual neurons and took the remembered and attended locations associated with the lowest combined rank as the population estimation. The estimated remembered location either agreed or disagreed with the actual remembered location of the selected trial, as did the estimated attended location in a separate computation. Repeating this procedure for a given number of neurons, n, more than 2,400 times—each time starting with a randomly selected set of test trials (more than 200 trials from each of the 12 conditions; four controls excluded)—yielded a percentage of correct estimations of the attended and remembered locations. We then calculated neuron-dropping curves for ensembles of size one to the total number of neurons, but typically the range 1–100 sufficed to capture the main features of the population estimation. To assess the representation of attended and remembered locations during the delays (see Figure 7 A– 7 D), we calculated neuron-dropping curves for the 800-ms period immediately preceding the onset of circle movement (gray curves in Figure 7 A– 7 D) and the 800-ms period immediately preceding the trigger signal (colored curves). Finally, we evaluated the time course of changes in these estimations, using neuron-dropping curves for a 200-ms window, which moved in 50-ms steps along the trigger-aligned records ( Figures 7 E, 7 F, and 8). The 200-ms window measured activity immediately before the time point plotted, to prevent the artifactual early appearance of a signal detection, and thus represents a “backward-looking” average. Supporting Information Figure S1 Rasters and Histograms from a Representative Attention Cell The activity matrix is the same as in Figure 3 A, measured in the 800 ms immediately prior to the trigger signal. This neuron is not the same as that illustrated in Figure 2 . Beneath the activity matrix, the rasters and histograms for each attended and remembered location are displayed in the format of Figure 2 A. (103 KB PPT). Click here for additional data file. Figure S2 Rasters and Histograms from a Representative Memory Cell The activity matrix is the same as in Figure 3 B, measured in the 800 ms prior to the trigger stimulus. Format as in Figure S1 . (95 KB PPT). Click here for additional data file. Figure S3 Rasters and Histograms from Two Representative Hybrid Cells The activity matrix in (A) is the same as in Figure 3 C, part a; the one in (B) is the same as in Figure 3 C, part b. Format as in Figure S1 . (151 KB PPT). Click here for additional data file. Figure S4 Activity Early Versus Late in the Delay Period A table of tuning indexes is given at the top for each of the cell classes (plotted in the bottom part of the figure), combinations of those classes, and other groups of cells as described in the left column. These population averages are divided into two groups of columns, those on the left showing data for the period before the circle began rotating (early) and those on the right showing data for the period after it had stopped and the monkey awaited the trigger signal (late). In the plot, the dashed line shows the median values, the dotted line shows the upper IQR. (56 KB PPT). Click here for additional data file. Figure S5 Activity Early Versus Late in the Delay Period Same PFdl neuron as in Figure 2 . The activity matrix in (C) comes from the data in (A), and the matrix in (D) comes from the data in (B), in the format of Figure 2 C. In (A), the red boxes enclose the measured period for the preferred location, 800 ms prior to the beginning of the circle's movement (200–1,000 ms after circle onset). In (D), the box shows the 800 ms immediately prior to the trigger stimulus. Note that the column-to-column variation in C necessarily results from chance variation because at that time the circle's final location is unknown. The figure shows, by example, that the spatial tuning in the period just before the triggering event strongly exceeds that before the circle begins moving, thus ruling out a strictly sensory account for spatial tuning (see also Figure S4 ). Note that after circle movement, responses to the circle were greater at the cell's preferred location (90°) but smaller at the least preferred location (270°). (188 KB PPT). Click here for additional data file. Figure S6 Activity Early Versus Late in the Delay Period Same PFdl neuron as in Figure S1 , in the format of Figure S5 . The red boxes show the measured period for the cell's preferred location in both (A) and (B). (156 KB PPT). Click here for additional data file. Figure S7 Neuron-Dropping Curves for the Two Monkeys Combined Format as in Figure 7 A– 7 D. (46 KB PPT). Click here for additional data file.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC524249.xml
534099	Comparison of frozen and RNALater solid tissue storage methods for use in RNA expression microarrays	Background Primary human tissues are an invaluable widely used tool for discovery of gene expression patterns which characterize disease states. Tissue processing methods remain unstandardized, leading to unanswered concerns of how to best store collected tissues and maintain reproducibility between laboratories. We subdivided uterine myometrial tissue specimens and stored split aliquots using the most common tissue processing methods (fresh, frozen, RNALater) before comparing quantitative RNA expression profiles on the Affymetrix U133 human expression array. Split samples and inclusion of duplicates within each processing group allowed us to undertake a formal genome-wide analysis comparing the magnitude of result variation contributed by sample source (different patients), processing protocol (fresh vs. frozen vs. 24 or 72 hours RNALater), and random background (duplicates). The dataset was randomly permuted to define a baseline pattern of ANOVA test statistic values against which the observed results could be interpreted. Results 14,639 of 22,283 genes were expressed in at least one sample. Patient subjects provided the greatest sources of variation in the mixed model ANOVA, with replicates and processing method the least. The magnitude of variation conferred by processing method (24 hours RNALater vs 72 hours RNALater vs. fresh vs frozen) was similar to the variability seen within replicates. Subset analysis of the test statistic according to gene functional class showed that the frequency of "outlier" ANOVA results within each functional class is overall no greater than expected by chance. Conclusions Ambient storage of tissues for 24 or 72 hours in RNALater did not contribute any systematic shift in quantitative RNA expression results relative to the alternatives of fresh or frozen tissue. This nontoxic preservative enables decentralized tissue collection for expression array analysis without a requirement for specialized equipment.	Background Many of the hopes for achieving clinical benefits of genomic medicine will hinge on the ability to develop an efficient specimen conduit from clinic to laboratory. Quantitative gene expression studies have created unprecedented tissue collection and handling challenges. In particular, the rapid degeneration of RNA, and possible perturbation of expression following excision place a high premium on prompt stabilization of tissue samples intended for expression analysis. This can be accomplished by sending a dedicated trained technologist outfitted with the necessary specialized equipment, such as liquid nitrogen, into the clinical environment. Alternatively, clinicians can be enabled to process the specimens directly in the course of patient care and send them in some stable form by unrushed and routine means for centralized processing. The latter is greatly preferred when patients are physically dispersed, and becomes essential in a multi-institutional setting. High throughput quantification of RNA expression in solid tissues has become a commonplace modality for genome-wide discovery of mechanisms of disease. Typically, groups of samples classified into comparison groups are used as a training set for expression pattern discovery, followed by validation in a fresh challenge set of annotated cases. The likelihood of success is highly dependent on the accuracy of classification within the training set, and ability to control random variables introduced during tissue processing and analytical measurement of RNA abundance. Efforts to standardize RNA quantification include sharing of information regarding probe design and use [ 1 ], or centralized design and production of analytical reagents and platforms by commercial entities using good manufacturing procedures (GMP). Flash freezing, either by immersion in liquid nitrogen or on dry ice, is the most common means of stabilizing tissue samples intended for RNA analysis. Local access to the necessary materials and expense of cold shipping and/or storage limit these collection capabilities in most clinical settings. An additional disadvantage of frozen storage is that homogenization of frozen tissue must be accomplished rapidly to avoid the rapid RNA degeneration that occurs during thawing of a previously frozen sample. Room temperature immersion of fresh tissue samples in aqueous sulfate salt solutions (such as ammonium sulfate) at controlled pH precipitates degenerative RNAses [ 2 ] and other solubilized proteins, thereby preserving the tissue with intact RNA [ 3 ]. Tissues preserved in this manner are compatible with most RNA isolation protocols, and may be archivally stored for extended periods at -60°C. A commercial preparation of this preservative, RNALater (Ambion), is increasingly being used by individual investigators and cooperative groups [ 4 ] for collection of human tissues. There have been promising reports of microarray-based RNA expression studies using RNALater-preserved tissues [ 5 - 10 ]. Solid tissues stored for a week in RNALater at room temperature give comparable RNA yields, and specific gene RNA abundance as with frozen tissue[ 8 ]. RNA yields are not affected substantially by storage at room temperature compared to 4°C, for storage intervals up to 3 months [ 11 ]. RNALater preserved tissues and cell suspensions are suitable starting points for RNA quantification by quantitative RT-PCR [ 11 ] and expression microarray hybridization [ 12 ]. One shortcoming of the prior work is that the potential changes contributed by RNALater use have not been precisely measured relative to random processing effects. We studied the effects of differences between storage conditions on gene expression as measured by expression array. Duplicate uterine myometrial tissue samples from three women were processed under each of 4 fixed storage conditions – fresh, frozen, 24 hours RNA-later and 72 hours RNA-later. The 24 labeled cRNA samples (Figure 1 ) were hybridized to HG-U133A Affymetrix microarrays. Then, for each microarray a data matrix was generated of 22,283 probe sets (genes) by quantitative expression levels in each RNA sample, and the effect of subject source, tissue processing, and replicates (Table 1 ) determined by ANOVA. Subset analysis by gene functional class was then performed to determine if storage condition has a specific effect on particular groups of genes. Figure 1 Experimental design. Tissue aliquots from 3 women were aliquoted, in duplicate, into four storage groups before RNA isolation and microarray hybridization. ANOVA design elements including fixed (storage group), random (woman, duplicate processing), and random interactive (woman × storage) effects as listed in Table 1. Table 1 Variability Sources in ANOVA Model (See Figure 1). Mixed ANOVA Model:Xij = u + ai + Bj + Eij where Xij is the observation (LN intensity), ai is the tissue storage effect, Bj is the individual variability effect and Eij is the noise term. Variability Source Type df Tissue Storage Fixed 3 Woman, Individual Variation Random 2 Interaction (Woman × Storage) Random 6 Replication (RNA isolation, chip Processing) Random 12 We found no systematic bias in measured quantitative level of gene expression by processing method, indicating that short term storage in RNALater is a valid alternative to traditional frozen storage. Results Of the 22,283 genes, 14,639 did not have absolutely null expression across all 24 samples. We fit the mixed model ANOVA from their log values and recorded this F statistic. The permutation distribution was used to assess the significance of F statistics calculated for each gene in the dataset. In this approach all 13,824 or (4!) 3 possible ways of permuting 4 pairs of replicate samples within each subject were considered. For each of these, the F statistics were computed for each gene. To control the overall error rate, the distributions of the maximum F statistics over the genes were used. That is, for each gene, the p-value is the proportion of permutations with the maximum F statistics over all genes greater or equal to the observed value for a particular gene. A test declaring as significant any genes with p < 0.05 then guarantees that the chance of any false positives being selected is < 5%. Similar analyses were performed replacing the distribution of the maximum F statistic with the distribution of the F statistics at the 95 th percentile and then at the 90 th percentile. After closer examination of the 387 genes in the 5% tail, we noted that most were exhibiting expression values below 100 for all 24 samples. In fact, within a storage condition, 2 out of 3 patients exhibited null expression while the third patient showed expression values other than null but less than 100 for at least one of their replicates. Therefore, as an additional analysis, any expression values less than 100 were recoded as 100. Genes that showed expression levels of 100 across all 24 samples, and therefore lacked variability, were then removed from the analysis. This resulted in 7,853 genes for which there was at least one sample with expression level greater than 100 across all 24 samples. Patient subjects provided the greatest sources of variation in the mixed model ANOVA, with replicates and processing method the least (Figure 2 ). The magnitude of variation conferred by processing method (24 hours RNALater vs 72 hours RNALater vs. fresh vs frozen) was similar to the variability seen within replicates. This is strong evidence that those individual expression profiles characteristic of the source tissue (woman) are unlikely to be obscured by the small amount of variation introduced by the processing method chosen. Figure 2 Sources of variation in the mixed ANOVA model . Distribution of Mean Squares Errors by variation source are plotted for those genes with at least one tissue showing expression at a level above LN(100) (7853 genes). Note that individual women emerge as the dominant source of variation. Variation contributed by tissue storage is of approximately the same magnitude as that seen between duplicate samples within the same storage group. Boxes encompass inner quartiles, horizontal line represents the median or the second quartile, and whiskers delimit 1.5 times the interquartile range. Because of the very large number of data points, outliers were suppressed in this summary plot. The distribution of ANOVA test statistics on those 7,853 genes where at least one sample of 24 had expression at a level exceeding 100 is compared in Figure 3 to those seen in the randomly permuted dataset. In the actual dataset, the maximum observed F statistic was 25.52; the observed F statistic at the 95th percentile was 3.51; and the observed F statistic at the 90 th percentile was 2.58. Corresponding p-values were 0.94, 0.55 and 0.51, respectively. The values of test statistics seen at the 95% level in a randomly permuted dataset (Figure 3 , thin solid line) are greater than those of the observed dataset (Figure 3 , thick solid line). This indicates that the model variation contributed by processing method is of the same magnitude as that seen randomly. Figure 3 Distribution of actual test statistics vs. randomly permuted background. Distribution of ANOVA F-statistics from the model shown in Figure 1 were calculated for the observed (FStat Observed) and permuted datasets. The maximum, 95th percentile, and 90th percentile F-Statistics in the permuted dataset provide an index of the distribution of test results expected for a random sample. The values of test statistics seen at the 95% level in a randomly permuted dataset are greater than those of the observed dataset. Genes were included if at least one tissue showed expression above LN(100) (7853 genes). Subset analysis of the test statistic according to gene functional class (Table 2 ) showed that the frequency of "outlier" ANOVA results within each functional class is overall no greater than expected by chance. Test statistic distribution was compiled by functional annotation for those 7853 genes which had at least one sample with detectable expression above a level of 100. Nine separate classification schemas containing a total of 127 functional classes were studied. Probe sets were rank ordered by decreasing ANOVA test statistic, and enrichment in the nominal p = 0.05 tail (F>3.51, containing top 387 of 7853 expressed) plotted by functional class (circles) and schema (box plot) (Figure 4 ). Table 2 Test statistic results by functional class within 9 annotation schema. Amongst 9 schema, a total of 127 functional classes ("Classes") contained expressed genes ("Total Genes"), and of these, 102 classes had a sufficient number of expressed genes (>50, "Class>50") to enumerate ("Genes F>3.51") and calculate fractional representation ("%>3.51", also Figure 3) in the nominal p = 0.05 tail at a test statistic cutoff of 3.51. 7853 genes with expression in at least one tissue >LN(100) were included. Code Schema Classes Class>50 TotalGenes Genes F>3.51 %>3.51 1 Biological Process (GO) 22 14 3591 197 5.5 2 Cellular Role (Proteome) 14 13 1775 92 5.2 3 Cellular Component (GO) 15 12 2845 154 5.4 4 Molecular Localization (Proteome) 9 10 1788 93 5.2 5 Organismal Role (Proteome) 17 12 1524 64 4.2 6 Biochemical Function (Proteome) 16 14 2727 150 5.5 7 Subcellular localization (Proteome) 8 7 2047 116 5.7 8 Molecular Function (GO) 21 15 4473 239 5.3 9 Pathways (GenMAPP) 6 5 756 33 4.4 Figure 4 Proportion of functional gene classes in which the test statistic falls within the nominal p < 0.05 tail. Nine publically available gene functional annotation schemas (listed in Table 2), each composed of multiple functional gene classes, were used to determine if specific functional subsets of expressed genes were more likely to show significant change in RNA detection between tissue treatments. Percentage of individual gene classes with test statistics above the nominal p = 0.05 level (F Statistic >3.51 in the dataset of 7853 expressed genes) are plotted on the Y axis for each of 9 classification schema (X axis). Results for functional classes of genes with a minimum of 50 available expressed genes are shown as individual data points (circles) the distribution of which is summarized for each schema by the superimposed notch plot. There are only two high outliers (arrows) amongst the 113 gene classes shown. These are a messenger RNA splicing factors in the Biological Process (GO) schema (Schema 1), and translation factor in the Pathways (GenMAPP) schema (Schema 9). This frequency of 2/113 outliers is no greater than expected by chance, (>5). Discussion Storage of fresh whole human tissues for up to 72 hours at room temperature in RNALater does not introduce quantitative bias into RNA expression determinations with the Affymetrix U133A array. Several differing standards justify this conclusion. First, by construction of a test model (Figure 1 ) incorporating both random reproducibility estimates (replicate determinations) and between-sample differences it has been possible to demonstrate that the magnitude of variation introduced by RNALater processing is equivalent to that seen within routine repeat specimens in a common processing group (Figure 2 ). Second, the extent of result variation conferred by RNALater processing is not statistically significant when measured against the randomly permuted dataset (Figure 3 ). This is an important element in evaluation of large datasets in which small numbers of individual variables may randomly demonstrate extreme values of the test statistic. Lastly, there is no evidence that specific functional subgroups of genes have aberrant behavior in this regard (Figure 4 ). Uterine myometrium was selected for these experiments because its components (myocytes, fibroblasts, vascular elements) are evenly intermingled throughout the myometrial compartment, lending itself to physical subdivision into equivalent aliquots. This would not be possible with more complex tissues in which differing cell types are distributed asymmetrically within the specimen. Despite the equivalency of subdivided fractions that underwent varying storage treatments, it must be noted that this is a hormonally responsive tissue whose expression patterns would be expected to differ between individual women as a function of monthly changes in circulating sex hormones. We did not control for hormonal factors or indication for hysterectomy (prolapse or fibroids) but selected patients randomly. It comes as no surprise that expression differences between women, irrespective of processing method, emerged as the dominant source of inter-sample variation. This was anticipated in constructing the model, by assigning the subject source of specimens as a random variable which could be measured against the fixed processing effects. It is likely that if a larger number of women had been included in the study, the observed biologic variation attributable to subject would have been even greater. Since our goal was to compare magnitude of variation contributed by subjects to that conferred by processing method, we achieved a balanced design by having comparable degrees of freedom for those two variables. There are several critical procedural elements that must be highlighted for successful preservation of solid tissues in aqueous sulfate salt solutions such as RNALater. These reagents enter the tissue through passive diffusion, a process which follows simple physical principles. The distance between the tissue surface, which is exposed to preservative, and the innermost regions of the fragment should be minimized. We did this by cutting the tissues into 2 mm thick slices, thereby reducing the diffusion distance to 1 mm or less. Clumping of multiple fragments into a mass that excludes preservative may obviate the benefits of fine division. This can be avoided either by gentle agitation or placement in a sufficiently large container that individual pieces are likely to disperse. Results reported here are for tissues stored at room temperature (23–25°C). Storage under cooler conditions (4°C) as recommend by the manufacturer of RNALater were not directly evaluated in this experiment because it was our intent to mimic storage interval and conditions commonly encountered when sending a specimen by express courier to a centralized processing facility. Storage at temperatures substantially higher than 25°C, especially before the preservative has had an opportunity to penetrate the tissue, should be avoided. Conclusions Split samples of fresh human tissue yield quantitatively similar RNA expression profiles whether processed fresh, frozen, or following 24–72 hour storage in RNALater. Formal statistical analysis shows patient source is the predominant source of variation between samples, with processing method contributing a random level of variation comparable to that seen in split duplicates (replicates). Subset analysis by functional gene category did not identify a specific class of genes which responded differently by processing method. Use of nontoxic ambient environment tissue preservatives makes it practical to engage practicing clinicians directly in decentralized sample collection for high throughput expression analysis in a central location. Tissue handling closely resembles that used by clinicians to prepare specimens for routine pathology analysis. Upon receipt in a centralized facility, the samples can either be immediately homogenized or archived at -60°C. Methods Tissue handling and storage Normal fresh uterine myometrial tissues were collected randomly from three women undergoing hysterectomy for benign uterine disease. For each hysterectomy, a single 4 to 8 gram tissue fragment was subdivided into eight aliquots composed of thin slices measuring no more than 2 mm in thickness. Replicate aliquots were immediately triaged into one of four storage conditions prior to homogenization: 1)immediate homogenization; 2)flash frozen in liquid nitrogen and storage for 48 hours at -80°C; 3)24 hour immersion in RNALater at room temperature with gentle agitation; or 4)72 hours immersion in RNALater at room temperature with gentle agitation. RNA isolation Tissue was solubilized in Trizol reagent (Gibco BRL, Grand Island, NY), and RNA isolated according to the manufacturers instructions. In brief, the aqueous phase was resolved by addition of chloroform, and RNA precipitated from the aqueous phase by addition of isopropyl alcohol. Pelleted RNA was washed with 70% ethanol, dried, and resuspended in water. Quality of total RNA was assessed by running a non-denaturing 1% agarose tris-acetate buffer which confirmed the integrity of 18S and 28S ribosomal bands for all 24 total RNA preparations. Microarray chip hybridization and data normalization Double-stranded cDNA was generated from 8 μg total RNA using the Superscript Choice System (Life Technologies) with T7-(dT)24 oligomer. cDNA was purified by phenol/chloroform extraction and ethanol precipitation. Biotin-labeled cRNA was prepared using the Enzo BioArray HighYield RNA Transcript labeling kit (Affymetrix). Unincorporated NTPs were removed from the biotinylated cRNA using an RNeasy kit (Qiagen). 10 μg of quality, fragmented cRNA was hybridized to each Affymetrix HG-U133A arrays containing probe sets representing approximately 22,000 genes. Array hybridization, washing was done according to the manufacturer's protocol (Affymetrix, GeneChip ® Expression Analysis Technical Manual) and all arrays were scanned under a low PMT (Photo Multiplier Tube) of 570 nm. Global scaling to a target value of 75 was applied to normalize all the arrays so they were comparable (Affymetrix Microarray Analysis Suite MAS5.0). The Affymetrix average-difference expression data and the P/A calls were used in the analysis. Those probe sets determined to have no detectible signal above background mismatch hybridization (Call of "Absent") were assigned a nominal value of 1 to facilitate future log transformations. Probesets having at least one tissue with detectable expression (call of "present") and an average difference above either 1 or 100 were selected to define subsets of 14639 permissively or 7853 stringently expressed genes, respectively. Further analysis was performed using the natural log transformed data of these probe subsets. Data files for all specimens processed are deposited online at the Gene Expression Omnibus at the National Center for Biotechnology Information [ 13 ]. Biostatistical analysis For this two factor study, a mixed model analysis of variance (ANOVA)was used, regarding storage condition as a fixed factor with four levels and subject as a random factor with three levels. The analysis of variance calculations for sums of squares in the mixed model ANOVA are identical to those for the fixed ANOVA model. Similarly, the degrees of freedom and mean squares are exactly the same. The mixed ANOVA model departs from the fixed ANOVA model only in the expected mean squares and the consequence choice of the appropriate test statistic. The mixed model also included a random storage by subject interaction. Replicate samples enabled us to estimate the replication error in the model. To test for the presence of storage main effects for each gene we divided the mean square for storage by the mean square for the interaction effect between storage and subject [ 14 ]. The ANOVA test statistic was calculated using C++. Functional annotation of probesets on the U133A chip, were downloaded from the Netaffx tm download center [ 1 ]. The March, 2003 version matches individual probesets with functional annotations (Table 2 ) from public domain databases including: the Gene Microarray Pathway Profiler, Gene Ontology Consortium, Proteome BioKnowledge Library, and Kyoto Encyclopedia of Genes and Chromosomes. Within each schema (comprised of many functional classes of genes), each gene is assigned to a primary functional class. Each probe set may be represented in several different schemas. Individual functional classes with at least 50 probesets represented within the U133A array were plotted by schema to show fractional representation within the nominal 0.05 tail (Figure 4 ). This provides a rapid and intuitive manner to identify functional classes of genes biased towards high test statistics in the ANOVA model. Authors' contributions GM and JW conceived and designed the research plan and participated in all aspects of data collection and analysis. DF participated in data analysis and interpretation. DN and DZ performed the statistical analysis. CL and HB performed the RNA isolations, chip hybridizations, and data collation.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC534099.xml
509318	A Red-Blooded Transcription Factor	null	Every multicellular organism depends on the coordinated actions of a multitude of cell types, each designed to carry out specific jobs. In vertebrates, for example, the task of ferrying oxygen to organs, tissues, and cells throughout the body is shouldered by red blood cells. These cells must develop early in the embryo to nourish the body and must be maintained at the right levels throughout adulthood to keep the organism healthy. Specialized red blood cells arise from undifferentiated stem cells in a developmental process called hematopoiesis. All vertebrates have the same basic plan for hematopoiesis during their development, with one group of stem cells producing embryonic blood cells and, later, another group in a different part of the body making adult blood cells. Such developmental programs are tightly orchestrated by suites of transcription factors—proteins that turn genes on and off. But which transcription factors act—and how and where they act—is still largely unknown. Scientists and doctors alike want to better understand differentiation in order to grasp not only how it proceeds normally but also how it can go wrong and lead to diseases, such as leukemia, in which white blood cells become cancerous, and aplastic anemia, in which bone marrow stem cells make too few red blood cells. To discover the genes involved in hematopoiesis, Leonard Zon's group at Children's Hospital, Boston, and their collaborators exposed zebrafish to mutagens and then studied the individuals that developed relevant disorders. In 1996, the group produced three zebrafish lines with embryos lacking fully differentiated red blood cells; since these fish also had especially shimmery tails, the lines were named moonshine. Now Zon and his colleagues have identified the mutant gene responsible for the moonshine zebrafish lines' peculiar traits. The moonshine (mon) gene, it turns out, encodes a transcription factor with wide effects on the embryo's mesoderm, the set of cells that eventually form the circulatory system, muscles, and skeleton. The researchers found that all three lines of mutant zebrafish had mutations in the mon gene. In the mutant fish, stem cells that produce red blood cells were present initially, but the defective moonshine protein was unable to keep the blood cells alive. These blood cells underwent apoptosis, or programmed cell death, leaving the fish unable to make red blood cells, and they died at about two weeks old. Hemoglobin staining shows that hematopoiesis is defective in a moonshine mutant (bottom) compared to a wild-type zebrafish embryo (top) The researchers found that the protein encoded by the mon gene is most similar to the human and mouse Tif1-gamma, one of a family of proteins known to link DNAbinding proteins with other factors that activate or repress gene activity. Using a DNA analysis technique in mouse cell cultures, the researchers found Tif1-gamma in nuclear bodies, multi-protein complexes in cell nuclei that help regulate gene expression. But the Tif1-gamma nuclear bodies didn't fit into any known class of such complexes, so it's an open question how Tif1-gamma acts and what other proteins help regulate it. Finding a transcription factor such as Tif1-gamma involved in early cell specialization opens a door to a suite of studies on the targets of the transcription factor, as well as other genes that act along with mon to affect red blood cell development.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC509318.xml
520749	Methylation profiles of thirty four promoter-CpG islands and concordant methylation behaviours of sixteen genes that may contribute to carcinogenesis of astrocytoma	Background Astrocytoma is a common aggressive intracranial tumor and presents a formidable challenge in the clinic. Association of altered DNA methylation patterns of the promoter CpG islands with the expression profile of cancer-related genes, has been found in many human tumors. Therefore, DNA methylation status as such may serve as an epigenetic biomarker for both diagnosis and prognosis of human tumors, including astrocytoma. Methods We used the methylation specific PCR in conjunction with sequencing verification to establish the methylation profile of the promoter CpG island of thirty four genes in astrocytoma tissues from fifty three patients (The WHO grading:. I: 14, II: 15, III: 12 and IV: 12 cases, respectively). In addition, compatible tissues (normal tissues distant from lesion) from three non-astrocytoma patients were included as the control. Results Seventeen genes ( ABL, APC, APAF1, BRCA1, CSPG2, DAPK1, hMLH1, LKB1, PTEN, p14 ARF , p15 INK4b , p27 KIP1 , p57 KIP2 , RASSF1C, RB1, SURVIVIN , and VHL ) displayed a uniformly unmethylated pattern in all the astrocytoma and non-astrocytoma tissues examined. However, the MAGEA1 gene that was inactivated and hypermethylated in non-astrocytoma tissues, was partially demethylated in 24.5% of the astrocytoma tissues (co-existence of the hypermethylated and demethylated alleles). Of the astrocytoma associated hypermethylated genes, the methylation pattern of the CDH13, cyclin a1, DBCCR1, EPO, MYOD1 , and p16 INK4a genes changed in no more than 5.66% (3/53) of astrocytoma tissues compared to non-astrocytoma controls, while the RASSF1A, p73, AR, MGMT, CDH1, OCT6,, MT1A, WT1 , and IRF7 genes were more frequently hypermethylated in 69.8%, 47.2%, 41.5%, 35.8%, 32%, 30.2%, 30.2%, 30.2% and 26.4% of astrocytoma tissues, respectively. Demethylation mediated inducible expression of the CDH13, MAGEA1, MGMT, p73 and RASSF1A genes was established in an astrocytoma cell line (U251), demonstrating that expression of these genes is likely regulated by DNA methylation. AR gene hypermethylation was found exclusively in female patients (22/27, 81%, 0/26, 0%, P < 0.001), while the IRF7 gene hypermethylation preferentially occurred in the male counterparts (11/26, 42.3% to 3/27, 11%, P < 0.05). Applying the mathematic method "the Discovery of Association Rules", we have identified groups consisting of up to three genes that more likely display the altered methylation patterns in concert in astrocytoma. Conclusions Of the thirty four genes examined, sixteen genes exhibited astrocytoma associated changes in the methylation profile. In addition to the possible pathological significance, the established concordant methylation profiles of the subsets consisting of two to three target genes may provide useful clues to the development of the useful prognostic as well as diagnostic assays for astrocytoma.	Background Diffusely infiltrating astrocytoma is a leading group of the primary central nervous system tumors, accounting for more than 60% of all primary brain tumors [ 1 , 2 ]. It may arise aggressively from the normal astrocytes, or evolve stepwise from the less its benign precursors. Owing to the difficulties with its early diagnosis and surgical removal of all residue diseased tissues, rapid progression, and frequent reoccurrence, the most advanced form of astrocytoma, glioblastoma (WHO grading IV) represents an extremely life-threatening intracranial malignant tumor both inside and outside of China [ 1 , 2 ]. Molecular genetic analyses have demonstrated multiple genetic lesions implicating to pathogenesis of astrocytoma, glioblastoma in particular. In addition to the frequent amplification and deletion of the EGF receptor gene ( EGFR ) [ 3 ], the main genetic events affecting the following tumor suppressor genes: the members of the INK4A initiated cell-cycle arrest pathway (the p16 INK4a ) [ 4 ], the p14 ARF [ 5 ], the RB1 [ 6 ] and the p53 [ 7 ]), a wide spectrum of the cell surface receptor genes (i.e., CD44 , integrin , and receptors for various growth factors), and the PTEN genes [ 8 ]. Transcription in eukaryotes is regulated at multiple levels and inversely correlated with the hypermethylated state as well as the chromatin condensation. It has been well established that the methylation status of CpG islands directly affects the DNA-protein interactions by eliminating the otherwise occurring sequence specific binding of the transcription factors whereas inducing the DNA-bindings of members of the methyl-CpG binding protein family (MBD). Histone modifications (deacetylation and methylation) may occur subsequently leading to chromatin condensation and a long-term transcriptional silencing status of the affected DNA segments. Over 40% of the protein coding genes have at least one CpG island within or near to their promoter, an strong indication for transcription of which is likely to be under the control of DNA methylation status. Three DNA methyl transferases are involved in the control of the methylation state of the CpGs in genome. DNA methyl transferase I is mainly responsible for the maintenance of the methylation status of the genome after DNA replication, while IIIA and IIIB act principally in the de novo DNA methylation in the early development of high eukaryotes. DNA methylation patterns in somatic cells are established during the early development and contribute to the allele-specific transcription silencing of the imprinted genes, including the silenced alleles in the X-chromosome and other chromosomes. The epigenetic pattern (the DNA methylation profiles of the genome) in high eukaryotes is integral to the normal execution of the biological activities in cells and needs to be actively maintained. In addition to the changes linked to the cell lineage specific pattern of gene expression, both global hypomethylation and local hypermethylation of the CpG islands occur progressively as cell ages. Aberrant DNA methylation pattern changes gene transcription that has been etiologically linked to cancer formation [ 9 , 10 ]. The genome-wide hypomethylation has been believed to activate transcription of the otherwise silenced transposon like repetitive sequences (such as the Alu and LINE repeats in mammals). As a result, the transposition occurs more prevalently so that the genomic instability in cancer cells will be significantly increased [ 11 - 13 ]. The hypermethylated state of the promoter CpG islands has been etiologically associated with transcription inactivation of a number of tumor suppressor genes in tumors, which are hypomethylated and transcribed in their normal counterparts. Therefore, the hypermethylated CpG island(s) of those genes have been regarded as a defect, reminiscent to the loss of heterozygosity or other types of genetic deletion for total inactivation of the tumor suppressor genes in cancer. The most noticeable example is the p16 INK4a gene that has been frequently hypermethylated in almost all types of the tumors examined [ 14 - 17 ] including hepatocellular carcinoma [ 18 ]. The loss of the genetic imprinting (changes in DNA methylation status) has been found to reactivate transcription of the otherwise silenced allele of the genes such as the insulin like growth factor 2 gene, which has been well documented in human tumors [ 19 ]. On the other hand, the reverse process, i.e., demethylation of the promoter CpG island, has also been found instrumental to the transcription activation of the otherwise inert genes in tumor cells [ 20 ]. A prominent example is the gene encoding the melanoma antigen, MAGEA1 that was hypermethylated and transcriptionally silenced in the normal liver tissues, and demethylated prevalently in the hepatocellular carcinoma tissues [ 18 ], correlating well with the elevated level of its expression in HCC [ 21 , 22 ]. The over-expressed gene, SURVIVIN , has also been reported to be demethylated in human ovarian cancer [ 23 ]. Despite of the fact that the elevated levels of expression of three DNA methyl transferase genes were detected in virtually all cancers, the profiles of the hypermethylated genes vary with both the types and stages of cancers. Therefore, the undefined defects in the epigenetic homeostasis during carcinogenesis, rather than the aberrant expression of any given DNA methyl transferase, are more likely to account for the cancer type specific pattern of DNA methylation at both global and local levels. Methylation profiling of the promoter CpG islands has been an important information gathering process for new insights into our understanding of the role of DNA methylation in both initiation and progression of human carcinogenesis. It would result in development of the DNA methylation based assays for cancer diagnosis as well as identification of the cancer genes suffering from the epigenetic defects . However, as the majority of studies had only targeted one or a few genes in rather small patient groups, the concurrent hypermethylation behavior of multiple genes has only been addressed in a limited number of tumor types, such as colorectal cancer. The majority, if not all, of the previous studies on the astrocytoma associated changes in methylation profiles have only examined a small number of genes for methylation status at the promoter CpG island [ 25 , 26 ]. In this study, we determined the methylation profiles of as many as thirty four genes in a cohort consisting of 53 astrocytoma patients and established the concordant methylation behavior of up to three targets. Our observations should provide new insights into the DNA methylation epigenetic defects in human astrocytoma. Methods All the experiments were performed according to protocols described previously [ 18 ]. The primer pairs for the methylation specific PCR were either adopted ( APC, BRCA1, CDH1, DAPK1, hMLH1, p14 ARF , p15 INK4b , RASSF1A, RB1 and VHL ) or designed according to the same principle with assistance of the software packages for the CpG islands identification and the primer design [ Additional file 1 ]. Tissue samples Tissue samples and DNA preparation With the informed consent of all patients and approval of the ethics committee, the tumor samples were collected from astrocytoma patients (n = 53) during operation at the Tiantan Neurosurgical Hospital in Beijing. The pathological classification of tumor tissues was carried out and the stage of each astrocytoma patients was determined according to the WHO classification [ 1 ]. No significant geographic impart was observed as patients came from different places in China and went to Beijing for treatment. In addition, the compatible tissues (normal tissues distant from the lesions) were surgically obtained from three non-astrocytoma patients [gangliocytoma (21 years old, male), angiocavernoma (49 years old, male) and meningioma (54 years old, female)] as the normal controls, which have been subjected to the proper pathological evaluation. Total genomic DNA was extracted from frozen tissue specimens (50 – 100 mg) according to a standard protocol with some modifications [ 18 , 27 ]. Frozen pulverized powders of the specimens were re-suspended with 2 ml lysis buffer: 50 mM Tris-HCl pH 8.0, 50 mM EDTA, 1% SDS, 10 mM NaCl plus 100 μg/ml boiling-treated RNase A (Sigma). Following one hour of incubation at 37°C, Proteinase K (Roche, USA) was added to the cellular lysates for a final concentration of 100 μg/ml and the digestion was carried out at 55°C for 2 hours. Organic extractions with a half volume of Phenol/Chloroform/Isoamyl alcohol (1:1:0.04) were repeatedly carried out until no visible interphase remained after centrifugation. DNA was precipitated from the aqueous phase in the presence of 0.3 M NaOAc pH 7.0 and two and a half volumes of ethanol and followed by one 70% ethanol-washing and dissolved at 65°C for 30 minutes with 0.2 – 0.4 ml TE (10 mM Tris-HCl pH 7.4 and 1 mM EDTA)and stored at 4°C till use. The DNA concentrations were calculated according to the OD 260 nm readings. Bisulphate treatment of DNA and Methylation specific PCR (MSP) The methylation status of the promoter CpG islands of thirty four genes in all DNA samples was analyzed by MSP on the sodium-bisulfite converted DNA [ 18 ]. In detail, 10 μg DNA in 50 μl TE was incubated with 5.5 μl of 3 M NaOH at 37°C for 10 minutes, followed by a 16 hour treatment at 50°C after adding 30 μl of freshly prepared 10 mM hydroquinone and 520 μl of freshly prepared 3.6 M sodium-bisulfite at pH 5.0. The DNA was desalted using a home-made dialysis system with 1% agarose (detailed protocol will be provided upon request). The DNA in the desalted sample (approximately 100 μl in volume) was denatured at 37°C for 15 minutes with 5.5 μl of 3 M NaOH followed by ethanol precipitation with 33 μl 8 M NH 4 OAC and 300 μl ethanol. After washing with 70% ethanol, the gently dried DNA pellet was dissolved with 30 μl TE at 65°C for 10 min. The DNA sample was finally stored at -20°C until further use. PCR reaction was carried out in a volume of 15 μl with 50 ng or less template DNA with FastStart Taq polymerase (Roche, Germany) as follows. After an initial heat denaturing step 4 minutes treatment at 94°C, 30 cycles of 92°C for 15 sec, varying temperatures with primer pairs ( Additional file 1 ) for 15 sec and 72°C for 20 sec, was carried out. The PCR products were separated by 1.2% ethidium bromide containing agarose gel electrophoresis with 1 × TAE and visualized under UV illumination. To verify the PCR results, representative bands from each target were gel-purified and cloned into T-vector (Promega, USA) followed by automatic DNA sequencing provided by BuoCai (Shanghai, China). Only verified results are presented in this report. To optimize the MSP procedure, the M. Sss I treated DNA was used as the methylated control template. In detail, the DNA from a normal liver tissue of the healthy liver donor [ 18 , 24 ] was batch cleaved with EcoR I, followed by M. Sss I treatment according to the manufacture's instruction (New England Biol., Boston, USA) for over night. The purified DNA was bisulphate treated as usual and subjected to MSP with the primer pairs for each of thirty three genes (except for the MAGEA1 gene), and only the verified targets were included for the study of the astrocytoma tissues. Statistical analysis The methylation data were dichotomized as 1 for the co-existence of the methylated and unmethylated alleles, 2 for methylated allele only and 0 for the unmethylated for both alleles to facilitate statistical analysis using contingency tables. The methylation profiles of each individual gene (in percentage) classified by the genders and grading of the patients were presented both in table and in plot. The statistic analyses for the association between the methylation profile of the gene and each of the clinical-pathological parameters were carried out with the statistics package , where both Pearsong's Chi-square test with Upton's adjustment and Fisher exact test were used to examine the tissue samples with the low expected values. The relative frequency with a 95% confidence interval (P < 0.05) for a binomial distribution was calculated for the whole as well as each subtype of astrocytoma patients. The concordant methylation behavior of the genes was established by comparing frequency of co-occurrence of 2 to 3 target subsets with a mathematic method, namely Discovery of Association Rules [ 28 ], which is frequently utilized for association analysis. Demethylation of U251 cells with 5-Aza-2'-deoxycytidine U251 cells (an established glioma cell line) were cultured in DEME plus 10% new born calf serum at 37°C in a 5% CO 2 atmosphere. When cell culture reached 50% confluence, they were treated with 5-Aza-2'-deoxycytidine (Sigma A3656) at the final concentration 10 and 20 nM, respectively for 3 days. The primer pairs for the RT-PCR (Table 1 ) was either adopted from published papers or designed with an assistance of the software . The total RNA was extracted with Trizol solution according to manufacturer's instruction (Invitrogen, USA), and cDNA was obtained using the Supertranscript plus reverse transcriptase with the oligo-dT as primers. PCR with single pair of the target primers run for 15 cycles, followed by another 15 cycle PCR reactions in the presence of the beta-actin primers (Table 1 ) (the parameter of each cycle is 94°C 20", 60°C for 20" and 72°C for 30"). The resulted PCR products were visualized under UV illumination after an electrophoretic separation on a 1.2% agarose. The methylation status of the target was analyzed by MSP. Table 1 The primers for RT-PCR analysis Primer Name sequence PCR Product Length (bp) Accession Number beta-actin L AAGTACTCCGTGTGGATCGG 616 NM_001101 beta-actin R TCAAGTTGGGGGACAAAAAG cdh13f GCTGGACTGGATGTTGGATT 246 NM_001257 cdh13t TTGAGGGTTGGTGTGGATTT magea1rf ACCTGACCCAGGCTCTGT 401 NM_004988 magea1rt CTCACTGGGTTGCCTCTG mgmtrf AAACGCACCACACTGGAC 404 NM_002412 imgmtrt AGGATGGGGACAGGATTG p73f AGATGAGCAGCAGCCACAG 218 NM_005427 p73t GTACTGCTCGGGGATCTTCA rassf1arf GTCTGCCTGGACTGTTGC 401 NM_007182 rassf1art AGCAGGGCCTCAATGACT Results and discussion Clinical-pathological classification To establish the methylation profile of thirty four genes during the process of astrocytoma development, we recruited 53 astrocytoma patients (27 female and 26 male; 49 primary and 4 recurrent) for this study. 14 cases were pathologically classified at the Grade I pilocytic astrocytoma (10–62 years old, mean: 39.1; 9 female, 5 male), 15 cases at the Grade II diffuse astrocytoma (4–50 years old, mean: 33.1; 10 female, 5 male), 12 cases at the Grade III anaplastic astrocytoma (1–72 years old; mean: 40.4; 4 female, 8 male), and 12 cases (including 4 recurrent cases) at the Grade IV glioblastoma (22–66 years old, mean: 44.6; SD = 22–66, 4 female, 8 male) (Table 2 ). The normal brain tissues distant from the lesions were also obtained from three non-astrocytoma patients who underwent brain surgery as normal controls in this study. Table 2 The clinical and pathological profiles of the patients Astrocytoma Non astrocytoma Gender female 27 1 male 26 2 Age, y <40 27 1 40–60 23 2 >60 3 0 Grade Age Mean Range I 39.1 10 to 62 14 II 33.1 4 to 50 15 III 40.4 1 to 72 12 IV 44.6 22 to 66 12 Recurrent 4 Primary 8 Aberrant Methylation profiling in astrocytoma The technical considerations The methylation-specific PCR (MSP) is widely used for methylation profiling of the genes in human cancers for both its easiness and sensitivity. However, the necessary steps have to be taken to eliminate both false positive and negative results. Comparing the MSP-data with the non-PCR data by Southern analysis of the methylation sensitive restriction enzyme is a valuable choice, as our previous work where the hypomethylated status of both p14 ARF and p15 INK4b genes shown by MSP was confirmed by Southern analysis [ 18 ]. Alternatively, the PCR reaction with the in vitro methylated genomic DNA (by M. Sss I) as template would be an ideal positive control for the absence of methylated targets in tumor tissue samples. By taking extra caution, we carried out MSP of all the targets with the M. Sss I treated normal liver DNA as positive control templates, except for the MAGEA1 gene was unmethylated in the normal liver tissue. While only the PCR reaction designated to the unmethylated template gave rise to the detectable bands with the parental DNA, the PCR bands were evident in both reactions with the M. Sss I treated DNA ( Additional File 2 ). Therefore, failure to detect the methylated alleles with the tissue samples should genuinely reflect the lack of methylated targets. To control the false positive with either pair of primers, the representative PCR products, were T-cloned and sequenced. Only the positive PCR results with the expected sequence profiles were scored and analyzed further. The methylation profiling of thirty four targets in astrocytoma Eleven of the thirty four target genes were previously studied either in astrocytoma or other types of tumors. The published PCR conditions for these genes: APC, BRCA1, CDH1, DAPK1, hMLH1, p14 ARF , p15 INK4b , p16 INK4a RASSF1A, RB1 and VHL ( Additional file 1 ) were adopted to enable the relevant inter-study comparisons if necessary. The remaining twenty three targets were selected from a list of genes displaying the altered pattern of the promoter CpG island in various biological settings including cancers. Their CpG islands were identified via bioinformatical tools and the primer pairs were designed accordingly [ 18 , 24 ]. Some of these thirty four genes have been shown to play a role in carcinogenesis, whereas the others have no obvious association with human carcinogenesis. Since it is still disputed whether DNA methylation mediated the gene silencing is causative in the malignant transformation of cell, we specifically selected both sets of genes in this study. The "cancer unrelated" genes selected encode erythropoiesis ( EPO ) [ 29 ], a ubiquitously expressed transcription factor ( OCT6 ) [ 30 ], and the myogenesis lineage-specific transcription factor ( MYOD1 ) [ 31 ]. The majority of the cancer associated genes examined were tumor suppressor genes including genes operating in the RB1/p16 INK4a pathway ( p14 ARF , p15 INK4b , p16 INK4a , and RB1 ) [ 32 ], and two cyclin-dependent kinase inhibitors ( p27 KIP1 [ 33 ] and p57 KIP2 ) [ 34 ]. Other genes in this subset were a p53 analogue:( p73 ) [ 33 , 35 ], two alternative forms of a tumor suppressors in the Ras mediated signal transduction pathway ( RASSF1A , and RASSF1C [ 36 ]), VHL [ 37 ], APC [ 38 ], PTEN [ 6 ], the deleted in bladder cancer chromosome region candidate 1 ( DBCCR1 ) [ 39 ], and the Wilms tumor 1 gene( WT1 ) [ 40 ]. We included the genes encoding the cell membrane proteins or nuclear receptors which act actively in the intercellular interactions: melanoma specific antigen A1 ( MAGEA1 ) [ 41 ], caveolin 1 ( CAV ) [ 42 ], chondroitin sulfate proteoglycan 2 ( CSPG2 ) [ 43 ], androgen receptor ( AR ) [ 44 ], and cadherins ( CDH1 [ 45 ] and CDH13 ) [ 46 ]. Three genes implicated in signal transduction were also selected: cyclin a1 [ 47 ], the interferon regulatory factor 7 ( IRF7 ), and a serine/threonine kinase 1 (Peutz-Jeghers syndrome) gene ( LKB1 ) [ 14 ]. There were the genes encoding the O-6-methylguanine-DNA methyltransferase ( MGMT ) [ 14 ]and metallothionein 1 A gene ( MT1A ) [ 48 ] which play a key role in the cellular response to alkalyting agents and heavy metal stress. The genes acting in DNA repair process were hMLH1 [ 49 ], and BRCA1 [ 50 ], while four genes are involved in apoptosis ( APAF1 [ 51 ], DAPK1 [ 15 ], and SURVIVIN [ 23 ]). Finally, the proto-oncogenes in this group were represented by v-abl homologue 1 ( ABL ) [ 52 ] ( Additional files 3 , 4 , 5 , 6 , 7 , 8 , 9 ). The genes displayed the uniformly unmethylated profiles in astrocytoma Of the unmethylated genes in all samples tested, EPO was a cancer unrelated gene, while "cancer associated" genes included ABL (1), APAF1 (2), APC (3), BRCA1 (5), CAV (6), CDH13 (8), DAPK1 (11), hMLH1 (14), LKB1 (16), p14 ARF (22), p15 INK4b (23), p27 KIP1 (25), p57 KIP2 (26), PTEN (28), RASSF1C (30), RB1 (31), SURVIVIN (32), and VHL (33) genes ( Additional files 3 , 4 , 5 , 6 , 7 , 8 , 9 ). Lack of hypermethylation of the RB1 gene in our observation was inconsistent with a recent report that the hypermethylated RB1 gene was detected in 19% of astrocytoma patients (26/136 cases analyzed) [ 53 ]. Since the same region was looked at in this work, the discrepancy noticed may simply reflect the inherent difference in the patient cohorts between our work and the published [ 53 ]. The genetic defects affecting the PTEN gene contributed to the pathogenesis of astrocytoma [ 54 ]. Lack of the hypermethylation of its promoter CpG island in both normal and astrocytoma tissues indicates that the DNA hypermethylation mediated silencing mechanism unlikely plays a significant role in the PTEN inactivation that occurs frequently in astrocytoma. This explanation might also be applicable to the no change type of methylation behavior for both the tumor associated genes ( ABL (1), APAF1 (2), APC (3), BRCA1 (5), CAV(6), CDH13 (8), DAPK1(11), hMLH1(14), LKB1(16), p14 ARF (22), p15 INK4b (23), p27 KIP1 (25), p57 KIP2 (26), PTEN (28), RASSF1C (30), RB1 (31), SURVIVIN (32), and VHL (33) genes) and the "cancer unrelated" genes ( EPO (14)) ( Additional files 3 , 4 , 5 , 6 , 7 , 8 , 9 ). The genes with the astrocytoma specific alteration in methylation As shown in Additional files 3 , 4 , 5 , 6 , 7 , 8 , 9 , thirteen genes ( CDH1 (7), CSPG2 (9), cyclin a1 (10), DBCCR1 (12), IRF7 (15), MGMT (18), MT1A (19), MYOD1 (20), OCT6 (21), p16 INK4a (24), p73 (27), RASSF1A (39) and WT1 (34)) were unmethylated in all three normal controls. In contrast, these genes were hypermethylated to various extents in the astrocytoma samples. The following six genes were marginally hypermethylated: p16 INK4a , EPO , DBCCR1 and MYOD1 genes were hypermethylated in 1.9% (1/53) of astrocytoma tissues, while both CDH13 and cyclin a1 genes were hypermethylated in 5.7% (3/53) of astrocytoma cases. No significant changes of these six genes shown in here acted against the notion that DNA methylation related mechanisms underline potential inactivation of this set of genes in the pathogenesis of astrocytoma. The infrequent hypermethylation of the p16 INK4a gene in astrocytoma was a total surprise, as it was frequently reported hypermethylated in various human tumors tested, including in HCC where we have previously found that the p16 INK4a , MYOD1 , CDH13 and cyclin a1 genes were frequently methylated [ 18 , 24 ]. To further verify this unexpected observation, we repeated the MSP analysis on five astrocytoma samples (shown unmethylated) along with one HCC sample (previously shown heterozygously methylated). As shown in panel 1, Fig. 1 , MSP patterns of the astrocytoma as well as HCC tissues remained the same. The identities of which were also confirmed by sequencing (panel 2, Fig. 1 ), showing that while the MSP products with the primers specific to the methylated targets in the HCC sample (Z92K) contained CpGs, the unmethylated targets in all the five astrocytoma tissues (21, 22, 26, A11 and B6) contained TpGs. Therefore, lack of hypermethylation of the p16 INK4a gene in astrocytoma was unlikely incorrect, which is consistent with a recent report that inactivation of the p16 INK4a gene in 48% of astrocytoma cases was genetic [ 55 ]. Figure 1 MSP/sequencing analyses of the p16 INK4a gene in astrocytoma and hepatocellular carcinoma Both electrophoretic patterns of the PCR products of the p16 INK4a in each of five astrocytoma cases (21, 22, 26, A11 and B6) and one HCC case (Z92K) (indicated respectively, at the top of figures) were presented. To indicate the methylation status, the sequenced data are aligned with the wild-type sequence. The remaining 7 targets were hypermethylated more frequently, occurring in 26.4% to 69.8% (14 to 37/53) of astrocytoma cases. The OCT6 gene was hypermethylated in 30.2% of the astrocytoma cases (16/53). Despite of the association of the OCT6 methylation with the aging process reported previously, we found no significant correlation/association of the OCT6 methylation to any clinical-pathological features, including age, gender and clinical grading of the patients. The significance of such a prevalent occurrence of the hypermethylated OCT6 gene remains to be determined. The RASSF1A (hypermethylated in 37/53 cases, 69.8%) is a variant of the recently identified tumor suppressor, the RASSF1 gene that acts at downstream of the Ras mediated apoptotic pathway and is capable of binding to Ras in a GTP dependent manner [ 36 ]. The RASSF1A gene has a more extended 5' part and its promoter CpG island displays a tumor specific hypermethylated profile in a variety of tumors, HCC in particular. Furthermore, lack of the RASSF1A expression in nineteen established tumor cell lines correlates with the hypermethylated state of its promoter CpG island [ 36 ]. The RASSF1C gene has its own promoter CpG island, but is not methylated in any tumors. The methylation behavior of these two genes was very similar to our previous observation in hepatocellular carcinoma, where 22/29 cases (79%) had the fully methylated 1A along with the unmethylated 1C variants [ 18 ]. As shown in Additional file 4 , 5 , 6 , 7 , 8 , 9 , the RASSF1A promoter-CpG island was methylated in 69.8% (37/53) of astrocytoma tissues, while the C variant was not methylated in any astrocytoma tissues. The hypermethylated state of the RASSF1A promoter CpG island was not correlated with gender, age and clinical grading. Consistent with the hypermethylated status of the RASSF1A gene in U251 cells, no expression at the mRNA level was detected. Partial demethylation of its promoter by the treatment with 5-Aza-2'-deoxycytidine indeed resulted in its transcription (Fig. 2 ). Figure 2 The methylation state and expression profiles of the CDH13, p73, MAGEA1, MGMT and RASSF1A genes in U251 astrocytoma cells before and after the demethylation treatment with 5-Aza-2'-deoxycytidine U251 cells were subjected to the 10 and 20 nM 5-Aza-2'-deoxycytidine (5-Aza) treatment for 3 days before both DNA and RNA were prepared for either MSP analyses or RT-PCR assessments. Panels; A, the methylation status of the CDH13, p73, MAGEA1, MGMT and RASSF1A genes and B, the expression profiles of each of these five genes, respectively in U251 cells. The p73 gene encodes a homologue to TP53, and loss of its heterozygosity has been observed in up to 90% of oligodendrogliomas and in 10–25% of diffuse astrocytoma [ 56 , 57 ]. In this study, we found that the p73 gene was prevalently methylated (25/53, 47.2%) with no significant association with any clinical-pathological parameters, such as gender and the WHO grading. The occurrence of the hypermethylated p73 gene was more prevalent in our results than a recent report which detected the hypermethylated p73 gene in 18% (5 /28) of the WHO grade IV but not in grade III astrocytoma [ 35 ]. Again, even the partially elevated demethylated status of its promoter CpG island in U251 cells resulted in reactivation of p73 transcription (Fig. 2 ). Both genetic defects and epigenetic abnormalities of the WT1 gene have been etiologically implicated in the formation of the Wilm's tumor [ 58 ]. In this study, we also found that the WT1 gene was hypermethylated in 30% (16/53) of cases, implying its possible involvement in the formation of astrocytoma. Tumor resistance to the cytotoxic chemotherapies may result from the disrupted apoptosis programs and remains a major obstacle in cancer treatment. In this study, the interferon regulatory factor 7 ( IRF7 ) gene was analyzed. The analogue ( IRF1 ) of IRF7 has been implicated in the IFN gamma mediated apoptosis with a profound effect on the chemo-sensitivity of tumor cells [ 59 , 60 ]. In consistence with the recent report that the IRF7 expression was negatively regulated by the promoter methylation [ 61 ], we found that the IRF7 gene was hypermethylated in astrocytoma (14/53, 26.5%) ( Additional file 4 , 5 , 6 , 7 , 8 , 9 ), with a strong male inclination (11/26, 42.3% verse the female group: 3/27, 11%, χ 2 = 6.632, P = 0.014). Although the gender difference remains to be understood, such a strong male association with IRF7 hypermethylation may have prognostic value. O(6)-methylguanine-DNA methyltransferase (MGMT), a DNA repair enzyme, removes alkylating adducts from the O(6) position of guanine and protects cells from cytotoxic and mutagenic stress. Silencing of the MGMT gene has been suggested to predispose the neoplastic clones to acquisition of the guanine to adenine point mutations in K- ras and p53 [ 62 ] and is associated with low-levels of micro-satellite instability in colorectal cancer [ 63 ]. We found that the MGMT gene was prevalently hypermethylated in astrocytoma (35%, 19/53), and its transcription could be reactivated by demethylation with 5-Aza-2'-deoxycytidine in U251 cells (Fig. 2 ). Hence, the MGMT hypermethylation in astrocytoma may indeed have the pathological significance. In this connection, a recent report suggested that the astrocytoma sensitivity to the alkylating type of chemotherapeutics might be contributed by the hypermethylated MGMT gene [ 64 ]. Expression of the metallothionein I A ( MT1A ) is inducible by a number of adversary agents such as heavy metals and oxidative stress. Both basal and inducible expression of this gene has been impaired in various tumor cell lines and attributed to the hypermethylated state of this gene [ 48 ]. In this study, we found that the MT1A gene was hypermethylated in 30% (16/53) of cases, with no significant gender and grading difference. The functional and pathological implications of the MT1A hypermethylation in astrocytoma remain to be established. Cadherins, the calcium-dependent proteins, contribute to various biological processes such as differentiation, migration and extra-cellular signal transduction of cell. Loss of expression of both E-cadherin ( CDH1 ) and H-cadherin ( CDH13 ) has been found in parallel with the hypermethylated promoter CpG islands in various cancers [ 65 , 66 ]. In this study, we found that the CDH1 gene was hypermethylated in 32.8% (17/53) of astrocytoma tissues, while the CDH13 gene was not methylated in all the astrocytoma tissues examined ( Additional files 4 , 5 , 6 , 7 , 8 , 9 ). In contrast, in human hepatocellular carcinoma [ 18 ], the CDH1 gene was unmethylated, while the CDH13 gene was frequently hypermethylated. Obviously, the molecular basis for tumor type specific methylation patterns of these two genes remains to be determined. Although the hypermethylation mediated gene silencing of the tumor suppressor genes is at the focal point of the epigenetic studies, demethylated status of the promoter CpG islands has been linked to the tumor associated activation of the normally silenced genes [ 19 - 23 ]. Therefore, we also studied both MAGEA1 and SURVIVIN genes. The promoter CpG islands were hypermethylated in normal tissues (for MAGEA1 in HCC [ 18 ] and for SURVIVIN in ovarian cancer [ 23 ]) and demethylated in parallel with the transcriptional activation in tumor cells. The unmethylated status of the SURVIVIN gene in astrocytoma is consistent with the over-expression of this gene (unpublished observations). However, its unmethylated status in all the non-astrocytoma tissues acts odd with the notion that its demethylation is associated with pathogenesis in human ovarian cancer reported previously [ 23 ]. Our previous studies indicated that demethylation of the promoter CpG island was correlated well with the over-expression profile of the MAGEA1 gene [ 18 , 21 ] in HCC. The MAGEA1 gene was fully hypermethylated in all four cases of the normal liver tissues but significantly demethylated in HCC tissues (21/28, 75%). It was found fully hypermethylated in all the three control tissues and in 74.5% (40/53) of the astrocytoma tissues and partially hypermethylated (13/53, 25.5%) in the other astrocytoma tissues. The occurrence of the MAGEA1 demethylation in HCC differed significantly from astrocytoma (75% verse 25.5%, P < 0.001). As it was fully methylated in the normal tissue, the partial hypermethylation (both hypermethylated and demethylated alleles existed) would imply that the event resulting in the loss of the hypermethylation state of the MAGEA1 gene indeed occurred in astrocytoma and should be scored positive for the changes in the methylation pattern in this study. The same principle has been applied for the opposite changes from the unmethylated pattern in the normal control to the partial or full hypermethylated status of all the other genes in astrocytoma tissues. It was also found that the partial demethylated status of the MAGEA1 gene in U251 cells induced by 5-Aza-2'-deoxycytidine occurred co-currently with activation of its transcription (Fig. 2 ). The gender association of the methylation profiles of the AR and IRF7 gene in astrocytoma By statistic analysis with both Pearson Chi-Square and Fisher's Exact tests, associations of the DNA methylation profiles of the targets displaying no less than 24.5% changes (the RASSF1A, p73, MGMT, CDH1, OCT6, WT1 as well as MAGEA1 genes) with the clinical pathological parameters (age, grading and gender) were assessed. The methylation profiles of the AR and IRF7 genes were found gender-oriented. The AR gene encodes the androgen receptor that plays a key role in the signal transduction pathways in response to the male steroid hormone, androgen and has been reported to be inactivated via the epigenetic mechanism in prostate cancers [ 67 ]. Physiologically, the AR gene should express exclusively in the somatic cells in males, while lacking of its expression in females is likely mediated by DNA methylation based mechanisms. Indeed, the hypermethylated along with the unmethylated AR genes were only found in the normal female brain tissue, but not from two male non-astrocytoma samples. The hypermethylation of the AR gene occurred frequently in the female group (81.5%, 22/27) but not in any males (0%, 0/26, χ 2 = 36.22, P = 0.000). It may simply be gender associated and do not have any significant relevance to carcinogenesis of astrocytoma. It was also noticed that hypermethylation of the IRF7 gene displayed an opposite gender inclination, detected in 11% of the female patients (3/27), and 42% of male patients (11/26, χ 2 = ?6.632, P = 0.014). Despite of the difficulty to offer a mechanistic interpretation, the potential prognostic value of such a gender-associated phenomenon might be worthwhile exploring in future. Demethylation by 5-Aza-2'-deoxycytidine treatment of the astrocytoma cells in culture resulted in partial demethylation and reactivated expression of the genes The hypermethylated status of the promoter CpG island has been linked to gene transcription silencing in a number of biological settings. The effect of the astrocytoma associated changes in the methylated state of the promoter CpG islands detected in this study on gene expression was assessed in U251 astrocytoma cells treated with the a demethylating agent, 5-Aza-2'-deoxycytidine. We used MSP to establish the methylation status of the promoter CpG island of all the genes with the astrocytoma associated methylation changes ( Additional files 3 , 4 , 5 , 6 , 7 , 8 , 9 ) in U251 astrocytoma cells, and analyzed the ability of 5-Aza-2'-deoxycytidine to demethylate five genes, as measured by MSP, and reactivate their expression, as detected by RT-PCR. As shown in panel 1 of Fig. 2 , while the CDH13 , MAGEA1 and p73 genes were heterozygously methylated, both MGMT and RASSF1A genes were fully hypermethylated in U251 cells. The CDH13 gene was found expressed, while the rest transcriptionally inert as measured by the RT-PCR. Although both methylated and unmethylated alleles for p73 and MGMT genes were evident in U251 cells, no expression was detected, indicating that the unmethylated allele may remain silent by the other mechanisms, including the genetic defects at critical control region. By the 5-Aza-2'-deoxycytidine treatment, both demethylation of the promoter CpG island and activation of transcription of these five genes were achieved (Fig. 2 ). Despite of the fact that demethylation of the promoter CpG islands was incomplete in samples treated with 20 nM 5-Aza-2'-deoxycytidine (Fig. 2 ), the expression of this five genes was either induced (the MAGEA1 , MGMT , p73 and RASSF1A genes) or elevated (the CDH13 gene). The concordant methylation behavior of the promoter CpG islands of the genes in Astrocytoma The DNA methylation mediated epigenetic changes also display the tumor type specific patterns, which seem to reflect the differentiation and maturation histories of the cell lineages as well as the aging process during which both global hypo- and local hyper-methylation occur. Hypermethylation of the promoter CpG islands in accord with the transcriptional silencing of the tumor suppressor genes, such as the p16 INK4a , and RASSF1A genes, has been well established in human tumors [ 16 , 68 ]. However, it remains unclear whether there is a common mechanism for the concurrent methylation changes of multiple tumor suppressor genes in tumors. To address this matter, it is necessary to examine a large number of genes for frequent changes in methylation in any type of human tumors. The concordant methylation behavior of multiple genes was firstly detected in colon cancer [ 69 ], based upon a comprehensive methylation profiling of over thirty genes. In this study, we have profiled the methylation status of thirty four genes in a cohort of 53 astrocytoma and 3 non-astrocytoma patients. Twenty three of these genes had not been studied previously in astrocytoma. As far as the number of the genes is concerned, this study is the most extensive in the astrocytoma field to our knowledge. Among thirty four genes, sixteen genes exhibited the astrocytoma associated changes in methylation profiles of the promoter CpG islands and nine genes displayed rather frequent changes (the occurrence ≥ 13/53, frequency ≥ 24.5%) ( Additional file 8 ). Four of 53 cases (7.5%) maintained the same methylation profile as the normal control. The rest 49 cases (92.5%) suffered from the methylation changes as much as no less than one target, an occurrence was significantly lower than in HCC, where all the cases displayed methylation changes affecting no less than three targets in the studies involved with twenty or twenty four targets [ 18 , 24 ], indicating that alterations in DNA methylation \occur more frequently in HCC than in astrocytoma. This may be contributed by the apparent anatomic inaccessibility of the brain to environmental adverse factors in comparison to the liver. The size of the subsets containing various number of the target affected (from one to nine) ranged from 1 to 11 cases, and peaked with 10 cases at three and 11 cases at five target subsets ( Additional file 9 ). To identify the most frequent changes of the target sets (one to three), a mathematic method called "the Discovery of Association Rules" [ 28 ] was used. The co-occurrence (case number/the total) and frequency (% of the total) of any subset of the targets that changed in methylation together in astrocytoma were counted and compared. In the entire cohort of patients in this study, the most altered target was the RASSF1A gene, 69.8% (37/53). The two genes that most altered together were the RASSF1A and p73 genes, hypermethylation of which was found in 20 (37.7%). Three genes that changed together were the former two plus CDH1 or OCT6 , hypermethylation of which occurred in 20.8% cases (11/53) (Column 2, a, Additional file 10 ). Furthermore, the occurrence in methylation change in any target in the two gene subsets was 79.3% (42/530 and in three gene subsets was 81.1–83% (43–44/53) (Column 3, a, Additional file 10 ). Since the hypermethylated AR is associated closely with the female gender of the astrocytoma patients and devoid of any association with the formation of astrocytoma, it was taken out from this analysis. Hypermethylation of the RASSF1A gene occurred in 21 female cases (77.8%, 21/27). Both RASSF1A and WT1 were hypermethylated in 13 (13/27, 48.1%); and the former two plus the hypermethylated p73 or CDH1 or OCT6 were found in 9 female cases (9/27, 33.3%), respectively (Column 2, b, Additional file 10 ). The subsets in the male patient group showed very different patterns. The single to three target subsets were the RASSF1A (16/26, 61.5%); the RASSF1A and IRF7 (10/26, 38.5%); and the former two plus the p73 or MGMT or MT1A (5/26, 19.2%), respectively (c, Additional file 10 ). In Grade I astrocytoma, the subsets for one, two and three targets were RASSF1A (10/14, 71.4%), RASSF1A plus p73 (6/14, 42.9%), and the former two plus either WT1 or IRF7 or MAGEA1 as well as RASSF1A plus CDH1 and WT1 (3/14, 21.4%). For Grade II astrocytoma, the corresponding sets consisted of the RASSF1A (12/15, 80%), the RASSF1A and MGMT or IRF7 (5/15, 33.3%), and the RASSF1A and MGMT plus p73 or OCT6 , or MT1A , or WT1 as well as the RASSF1A and IRF7 and MT1A (3/15, 20%), respectively. For Grading III astrocytoma, those subsets were composed of the RASSF1A (8/12, 66.7%), the RASSF1A and CDH1 (5/12, 41.7%), and the formal two plus MGMT (4/12, 33.3%), respectively. For Grading IV astrocytoma, the comparative subsets contained the RASSF1A or p73 (7/12, 58.3%), the RASSF1A and p73 (6/12, 50%), and the former two plus MGMT or OCT6 (4/12, 33.3%), respectively. (d-g, Additional file 10 ). Our methylation profiling efforts described in this report provided the following informative targets: the RASSF1A , p73 , WT1 , MGMT , CDH1 , OCT6 , and IRF7 genes. The established concordant methylation profiles of these eight genes ( Additional file 10 ) may provide useful clues for the epigenetic biomarker selection to for the novel diagnostic and prognostic assays of astrocytoma. The hypermethylated status of this lest of genes in the serum, and biopsies of the suspected astrocytoma patients may serve as good diagnostic indicators, if they can be reliably detected. With the death/survival profiles of this cohort of astrocytoma patients available in the future, the methylation profile established in this study may have certain prognostic value. Abbreviations HCC: Hepatocellular carcinoma; PCR: polymerase chain reaction; MSP: methylation specific PCR; ABL : v-abl Abelson murine leukemia viral oncogene homolog 1; APAF1 : apoptotic protease activating factor; APC : adenomatosis polyposis coli; AR : androgen receptor; BRCA1 : breast cancer 1; CAV : caveolin 1, caveolae protein; CDH1 : cadherin type 1, E-cadherin; CDH13 : cadherin 13, H-cadherin; CSPG2 : chondroitin sulfate proteoglycan 2 (versican); cyclin a1 : cyclin A1; DAPK1 : death-associated protein kinase 1; DBCCR1 : deleted in bladder cancer chromosome region candidate 1; EPO : erythropoietin; hMLH1 : mutL homolog 1, colon cancer, nonpolyposis type 2; IRF7 : interferon regulatory factor 7; LKB1 : serine/threonine kinase 11 (Peutz-Jeghers syndrome); MAGEA1 : melanoma antigen, family A, 1 (directs expression of antigen MZ2-E); MGMT : O-6-methylguanine-DNA methyltransferase; MT1A : metallothionein 1A (functional); MYOD1 : myogenic factor 3; OCT6 : POU domain, class 3, transcription factor 1; p14 ARF : the alternative reading frame of the cyclin-dependent kinase inhibitor 4a; p15 INK4b : cyclin-dependent kinase inhibitor 4b; p16 INK4a : cyclin-dependent kinase inhibitor 4a; p27 KIP1 : cyclin-dependent kinase inhibitor 1B (p27, KIP1); p57 KIP 2 : cyclin-dependent kinase inhibitor 1C (p57, KIP2); p73 : tumor protein p73; PTEN : phosphatase and tensin homolog; RASSF1A : ras association (RalGDS/AF-6) domain family 1 protein isoform 1a; RASSF1C : ras association (RalGDS/AF-6) domain family 1 protein isoform 1c; RB1: retinoblastoma 1; VHL : von Hippel-Lindau syndrome; WT1 : Wilms tumor 1. Competing interests None declared. Authors' contributions JY, HYZ, JG, executing the experiments; SL and JHL, providing the patient samples; WL and YFW, carrying out the mathematic analyses of the data JDZ: designing and organizing experiments as well as completing manuscript. Pre-publication history The pre-publication history for this paper can be accessed here: Supplementary Material Additional File 1 The target promoter CpG islands and the primer pairs for methylation specific PCR. This file contains his study. Click here for file Additional File 2 Methylation profiles of thirty three genes on the in vitro methylated genomic DNA by M. Sss I methyl transferase. The Eco RI restricted genomic DNA from the liver tissue of a healthy donor was in vitro methylated overnight with M. Sss I methyl transferase and subjected to the MSP analysis, followed by electrophoresis in a 1.3% agarose gel. , the DNA size markers, NL, the untreated sample, U and M, MSP with the pair of primers specific to the unmethylated and methylated targets, respectively. Panels: 1, ABL; 2, APAF1; 3, APC; 4, AR; 5, BRCA1; 6, CAV; 7, CDH1; 8, CDH13; 9, CSPG2; 10, cyclin a1; 11, DAPK1; 12, DBCCR1; 13, EPO; 14, hMLH1; 15, IRF7; 16, LKB1; 17, MGMT; 18, MT1A; 19, MYOD1; 20, OCT6; 21, p14 ARF ; 22, p15 INK4b ; 23, p16 INK4a ; 24, p27 KIP1 ; 25, p57 KIP2 ; 26, p73; 27, PTEN; 28, RASSF1A; 29, RASSF1C; 30, RB1; 31, SURVIVIN; 32, VHL and 33, WT1. Click here for file Additional File 3 Methylation profiles of thirty four genes in astrocytoma (part I). Both electrophoretic patterns of the representative PCR products of each of thirty four targets (indicated respectively, at the top of figures) and the sequencing verification of the one representative PCR product were presented. To indicate the methylation status, the sequenced data are aligned with the wild-type sequence. , size markers, the bands of 250 bp and 100 bp were shown. U, the unmethylated; M, the hypermethylated. Panels: 1, ABL; 2, APAF1; 3, APC; 4, AR; 5, BRCA1; 6, CAV; 7, CDH1; 8, CDH13; 9, CSPG2; 10, cyclin a1; 11, DAPK1 and 12, DBCCR1. Click here for file Additional File 4 Methylation profiles of the promoter CpG islands of thirty four genes in astrocytoma (part II). Both electrophoretic patterns of the representative PCR products of each of thirty four targets (indicated respectively, at the top of figures) and the sequencing verification of the one representative PCR product were presented. To indicate the methylation status, the sequenced data are aligned with the wild-type sequence. , size markers, the bands of 250 bp and 100 bp were shown. U, the unmethylated; M, the hypermethylated. Panels: 13, EPO; 14, hMLH1; 15, IRF7; 16, LKB1; 17, MAGEA1; 18, MGMT; 19, MT1A; 20, MYOD1; 21, OCT6 and 22, p14 ARF> . Click here for file Additional File 5 Methylation profiles of the promoter CpG islands of thirty four genes in astrocytoma (part III). Both electrophoretic patterns of the representative PCR products of each of thirty four targets (indicated respectively, at the top of figures) and the sequencing verification of the one representative PCR product were presented. To indicate the methylation status, the sequenced data are aligned with the wild-type sequence. , size markers, the bands of 250 bp and 100 bp were shown. U, the unmethylated; M, the hypermethylated. Panels: 23, p15 INK4b ; 24, p16 INK4a ; 25, p27 KIP1 ; 26, p57 KIP2 ; 27, p73; 28, PTEN; 29, RASSF1A; 30, RASSF1C; 31, RB1; 32, SURVIVIN; 33, VHL and 34, WT1. Click here for file Additional File 6 The summary of the astrocytoma cases displaying no or changes in the methylation profiles (part I). The frequency (%) of the astrocytoma displaying no or the changes in the methylation profile of each target from the normal control were counted and presented in table as well as plotted in the figure below. The filled, shading and empty boxes indicate the cases where only hypermethylated allele, both hypermethylated and unmethylated alleles and only unmethylated alleles were detected, respectively. The frequency (%) of the hypermethylated targets (except for the MAGEA1 gene) among the total cases was scored for positive changes in astrocytoma. The MAGEA1 was fully methylated (3/3, 100%) in the control, and become partially demethylated in some astrocytoma, therefore, demethylation of the MAGEA1 in astrocytoma was scored as positive changes. Sub-tables: a, the female patient group, b, the male patient group, and c, the control. Click here for file Additional File 7 The summary of the astrocytoma cases displaying no or changes in the methylation profiles (part II). The frequency (%) of the astrocytoma displaying no or the changes in the methylation profile of each target from the normal control were counted and presented in table as well as plotted in the figure below. Sub-tables d-h, the WHO grading I to IV, respectively; The filled, shading and empty boxes indicate the cases where only hypermethylated allele, both hypermethylated and unmethylated alleles and only unmethylated alleles were detected, respectively. The frequency (%) of the hypermethylated targets (except for the MAGE A 1, where the heterozygously hypermethylated) among the total cases was presented in the plot. Click here for file Additional File 8 The occurrences and frequency of changes in methylation. , One of three cases was methylated; *, The MAGEA1 gene was fully methylated in the normal tissues and partially demethylated in astrocytoma patients as indicated in the relevant cells. Therefore, the astrocytoma associated changes in methylation of this gene is opposite to the rest, i.e., demethylation rather than hypermethylation. Figure is each cells are the frequency in % and occurrence (case number). Click here for file Additional File 9 The summary of changes in the methylation pattern in subsets. Both occurrence (case number) and frequency (%) for the subsets having no change in methylation and changes in one to nine genes are presented in % and (case number) in the top half of table, which was also plotted. Both occurrence (case number) and frequency (%) for the subsets having no change in methylation and changes in, at least, one to nine genes are presented in % and (case number) in the bottom half of table. Click here for file Additional File 10 The summary of the concordant methylation behavior of the hypermethylated targets in astrocytoma. The co-occurrence (/total case) and frequency (%) of a panel subsets consisting of one to three targets were treated with method "Discovery Association Rules" and presented. Sub-tables: a, the total, b, the female, c, the male, and d-g, the grade I to IV, respectively. Column 1 is the number of target in each subset. Column 2 is the co-occurrence (case number/total) (frequency in %). Column 3 is the occurrence of any single target in each subsets, presented in case number (frequency %). The column 4 is the gene(s) in subset. N.B., In view of the strong female inclination of the AR methylation and lacking of any association with astrocytoma, AR has been taken off from this analyses. Click here for file	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC520749.xml
522823	Spatial dependency of Buruli ulcer prevalence on arsenic-enriched domains in Amansie West District, Ghana: implications for arsenic mediation in Mycobacterium ulcerans infection	Background In 1998, the World Health Organization recognized Buruli ulcer (BU), a human skin disease caused by Mycobacterium ulcerans (MU), as the third most prevalent mycobacterial disease. In Ghana, there have been more than 2000 reported cases in the last ten years; outbreaks have occurred in at least 90 of its 110 administrative districts. In one of the worst affected districts, Amansie West, there are arsenic-enriched surface environments resulting from the oxidation of arsenic-bearing minerals, occurring naturally in mineral deposits. Results Proximity analysis, carried out to determine spatial relationships between BU-affected areas and arsenic-enriched farmlands and arsenic-enriched drainage channels in the Amansie West District, showed that mean BU prevalence in settlements along arsenic-enriched drainages and within arsenic-enriched farmlands is greater than elsewhere. Furthermore, mean BU prevalence is greater along arsenic-enriched drainages than within arsenic-enriched farmlands. Conclusion The results suggest that arsenic in the environment may play a contributory role in MU infection.	Background Buruli ulcer (BU) is a skin disease, which usually begins as a painless nodule or papule and may progress to massive skin ulceration. If untreated BU may lead to extensive soft tissue destruction, with inflammation extending to deep fascia. The parts of the body most affected are the extremities. Subsequent complications may include contractural deformities. The main form of treatment is wide excisional surgery, including amputation of limbs, which requires prolonged hospitalization and is thus a significant burden on hospital resources and budgets. In recent years, there has been increased incidence of BU in West Africa (including Benin, Burkina Faso, Cote d'Ivoire, Ghana, Guinea, Liberia and Togo), Mexico, French Guyana, Papua New Guinea and Australia. The disease seems to affect mostly impoverished inhabitants in remote and rural areas; children are the most vulnerable, accounting for about 70% of the cases [ 1 ]. The World Health Organization (WHO) has recognized BU as the third most prevalent mycobacterial disease after tuberculosis and leprosy and has called for urgent action to control it [ 2 ]. The causative agent of BU is Mycobacterium ulcerans (MU), which was first described in Bainsdale, Australia, in 1948 [ 3 ]. From the medical point of view, MU is among the group of mycobacteria that are potentially pathogenic in humans and animals under special circumstances [ 4 ]. It is suggested MU enters through a small break or trauma in the skin because it is not known to penetrate through intact or healthy skin [ 5 , 6 ]. Portaels et al. [ 7 ] have suggested that insects may be involved in the transmission of the disease because insects found in the roots of trees tested positive with the mycobacterium. Marsolliers et al. [ 8 ] found through an experimental study that the bite of MU-infected waterborne insects transmitted infection to mice. In terms of human infection, however, the reservoir of MU and the mode of transmission of BU are still unclear [ 4 , 9 - 12 ]. Epidemiological data suggest that environmental factors such as climate, soil, geology, geochemistry, etc. may indirectly influence or contribute to MU infection [ 4 ]. In addition, the frequencies of some diseases caused by mycobacteria indicate that species are distributed geographically [ 13 ]. For example, MU has been observed mainly in the tropics and [ 4 ] especially in anthropogenically-polluted areas [ 14 ]. Since MU is known to be present in nature although its reservoir is not known and since the epidemiology of BU is still unclear, there is a need to have a better understanding of environmental, ecological, and behavioural factors that predispose to infection. Spatial analysis potentially contributes important information leading to the understanding of the epidemiology and etiology of BU. The main objective of this paper is to explore relationships between some spatial environmental factors and the prevalence of BU. Results For the buffers tested, P values range from 0.09 to 0.46. The buffers with the highest P values (i.e., 0.09) are 100 m for drainage channels and 400 m for farmlands. With these buffers 24 of the 61 settlements (i.e., 39%) fall within 100 m of arsenic-enriched drainage channels (Figure 4 ) and 41 of the 61 settlements (i.e., 67%) fall within 400 m of As-enriched farmlands (Figure 5 ). The mean BU prevalence within the drainage buffer is 0.7% whereas the mean BU prevalence inside the farmland buffer is 0.6%. Thus the naturally smaller number of settlements within the drainage buffer (i.e., 24) has a slightly higher BU prevalence than the relatively larger number of settlements within the farmland buffer (i.e., 41). Figure 4 Map of As-enriched farmlands. Distances to As-enriched farmlands (red; with > 15 ppm As in stream sediments) and locations of villages with BU cases. Figure 5 Map of As-enriched streams. Distances to As-enriched streams (black; with > 15 ppm As in stream sediments) and locations of villages with BU cases. Discussion Siting of rural settlements in the study area is based primarily on proximity to and availability of water for drinking and other domestic purposes. Consequently, many settlements are located within the optimum buffer distance of 100 m from drainage channels. Where water is abstracted from drainage channels enriched in arsenic, chronic ingestion of arsenic-enriched water through drinking and cooking is likely. This renders the inhabitants susceptible to several kinds of diseases [ 15 , 16 ] including BU. Amofah et al. [ 17 ] studied 90 BU patients and found that 52 used surface water as the source of their drinking water. The result of the statistical analysis corroborates this observation in that BU prevalence is highest where the inhabitants have ready access to domestic water supplies from arsenic-enriched surface drainage. Subsistence farmlands, especially those that depend partially on irrigation, tend to be located along stream floodplains. Soils in these floodplains have a high cation exchange capacity [ 18 ] so that, where streams carry high concentration of arsenic, there is accumulation of arsenic in the soils of the floodplains. These high concentrations of arsenic are in part taken up by the foodcrops grown there [ 19 - 21 ]. The results of the statistical analysis suggest that a high proportion of settlements with high BU prevalence exploit such floodplain farmlands enriched in arsenic. Through consumption of arsenic-enriched drinking water and arsenic-enriched foodcrops, inhabitants in some settlements in the Amansie West District are prone to chronic ingestion of higher-than-average (but sub-toxic) levels of arsenic. Arsenic interacts with and inhibits several enzymes in the body [ 15 ] leading to several multisystemic non-cancer effects [ 16 ], which could predispose to defect the immune system [ 22 ]. Subjects exposed to high levels of arsenic concentrations were known to have impaired immune response [ 23 ]. Immunosuppression due to arsenic has been found to defect antigen processing of splenic macrophages with consequent defective mechanism of helper T-cells [ 24 , 25 ]. Down-regulation of the immune system is known to be a risk factor for the development of BU [ 26 , 27 ]. Several studies [e.g., [ 28 - 31 ]] have reported of impaired resistance to viral/bacterial infection via arsenic ingestion. Conclusions The results of this study reveal spatial dependency of BU prevalence upon proximity to drainage channels and farmlands containing > 15 ppm arsenic. Proximity implies chronic exposure to and/or ingestion of elevated concentrations of arsenic, which influences susceptibility to infection. Methods Research hypotheses It has been consistently theorized that BU is acquired when MU enters the body through a skin rupture [ 32 , 27 ]. However, several people who were affected by the disease do not recall having any break or trauma in their skin prior to being infected [ 33 ]. A possible alternative is entry through non-ruptured but unusually unhealthy or thin skin. Several dermatological diseases (e.g., Bowen's disease, hyperkeratosis, hyperpigmentation) are related to arsenic ingestion and exposure [ 34 ]. Bioaccumulation of arsenic in the fatty tissues of the skin [ 35 ], due to its high lipid solubility [ 36 , 37 ], may provide a favourable environment for MU in the skin because arsenic is known to help microorganisms grow [ 38 ]. It can be hypothesized, therefore, that (a) arsenic induces MU adhesion to human tissues and (b) arsenic influences the ability of MU to establish BU. In a case study, Amofah et al. [ 17 ] reported that about 44% of the BU patients were farmers whilst about 54% were school children. In Ghana many children help their parents on farms. Not only do farmers and children come in contact with natural drainage areas on their journeys to and from their farmlands, but also the farms are located near water bodies or drainage systems for obvious irrigation purposes [ 39 ]. If farmlands and surface drainage channels are contributory factors to BU, farmlands and surface drainage channels enriched in arsenic may contribute to still higher prevalence of BU. Research methodology Spatial analysis of data provides opportunities for epidemiologists to study associations between environmental factors and spatial distribution of diseases [ 40 ]. A geographic information system (GIS) is capable of analyzing and integrating large quantities of geographically distributed data as well as linking geographic data to non-geographic data to generate information useful in further scientific (or medical) research and in decision-making. In this study, topographic map data, stream sediment geochemical data for arsenic, ASTER satellite imagery and locations of settlements with BU cases were the basic data inputs into the GIS. Spatial data processing was carried out (a) to delineate arsenic-enriched catchment basins based on arsenic concentrations in stream sediment samples, (b) to delineate farmlands from ASTER satellite imagery and determine arsenic-enriched farmlands based on catchment basin data and (c) to extract drainage channels from the topographic map and determine arsenic-enriched drainage channels based on arsenic-enriched catchment basins. Proximity analysis was undertaken to determine spatial relationships between BU-affected areas and the arsenic-enriched areas determined from the data inputs. The study area History of BU in Ghana The study area is in Ghana, where the first case of BU was reported in 1971 and, between 1991 and 1997, more than 2000 cases have been reported [ 41 ]. The disease has affected all of the ten regions and at least 90 of the 110 districts in Ghana [ 42 ]. The Ashanti Region is the worst affected, accounting for about 60% of all reported cases, of which the greatest percentage is in the Amansie West District (Figure 1 ). Figure 1 The study area. Amansie West District, Ghana, showing the study area (box) and villages with BU cases (black dots). Location of study area The Amansie West District lies between latitudes 6°N and 6°45'N and longitudes 1°30'W and 2°15'W. It covers an area of about 1,136 km 2 . The district capital, Manso Nkwanta, is about 40 km south of Kumasi. The district is drained by the Offin and Oda rivers. Vegetation in the district is composed mainly of secondary forests, thicket, forb regrowth (i.e., soft-stemmed leafy herbs, mostly the weeds, which appear on farms and have to be cut regularly) and swamp vegetation. Vegetation thrives in ferric fluvisols, which are the major soil types in the district. These soils have been developed through yearly rainfall ranging from 125 to 200 cm with temperatures of 22 to 30°C. The landscape of the district varies from gentle to broken. The district is underlain by Lower Proterozoic Birimian and, to a lesser extent, Tarkwaian rocks. Throughout Ghana, Birimian rocks of West Africa are mainly volcanic greenstones with intervening sedimentary rocks and granitoid intrusions, in places containing deposits composed of pyrite, arsenopyrite, minor chalcopyrite, sphalerite, galena, native gold and secondary hematite [ 43 ]. The district has about 310 settlements (though not all these settlements are mapped) with a population in 2000 of 108,726. There are approximately equal percentages of males and females (49% and 51%, respectively), of whom 70% are farmers and 22% are engaged in legal and 'galamsey' (or illegal) mining. The study area is the east-central part of the Amansie West District (covered by a single topographic map sheet, 0602C1), with an area of about 623 km 2 , comprising 61 settlements and including the Bilpraw goldmine (formerly a treasure mine of the Ashanti Kings). The BU cases per settlement range from 1–29. Materials The following are the sources of spatial data input to the GIS. • Incidence of BU per settlement in 1999, obtained from Korle-BU Teaching Hospital, Accra, Ghana. • Settlement population estimates for 2000, projected by the Ministry of local government and rural development. • Topographic map (Sheet 0602C1, 1974, at a scale of 1: 50,000), a single sheet covering the study area, obtained from the Survey Department, Accra, Ghana. • Location map (at scale of 1: 62,500) of stream sediment samples collected in part of the Amansie West District in 1992 and list of arsenic concentrations determined in these samples, obtained from the Geological Survey Department, Accra, Ghana. • Boundary map (at scale of 1: 250,000, surveyed in 1991) of the district, obtained from the Amansie West District Administration. • ASTER imagery (level 1B) acquired on 15/01/2002, obtained from the US Geological Survey. • Landuse/landcover map of Ghana (traced on Landsat TM data of 1998 and published in the same year), obtained from the Remote Sensing Application Unit (RSAU), University of Ghana, Legon. The GIS operations were carried out in three principal steps: (1) spatial data capture; (2) generation of spatial factor maps; and (c) spatial data analysis. The GIS operations were carried out using ILWIS (Integrated Land and Water Information Systems), a GIS software package developed by the International Institute for Geo-information Science and Earth Observation (ITC) in the Netherlands. Spatial data capture The different analog maps were scanned then georeferenced (by defining the x and y coordinates of the corner points of the maps) into a UTM coordinate system. From the scanned map, spatial data were captured by screen digitizing. From the topographic map, rivers, streams and gullies were digitised as line segments as were elevation contours. The boundaries of the district were digitised as line segments and then polygonized. The locations of centres of 61 settlements (identifiable on the topographic map) were digitised as points and the BU incidence in 1999 was recorded as spatial attribute of each settlement. From the stream sediment sample location map, the locations of the samples were digitised as points and the arsenic concentrations (in ppm) were recorded as a spatial attribute of each sample. The ASTER imagery was also georeferenced to the same coordinate system using eight reference points (tie points), which were selected in the image and which could be identified in the topographic map. Using an affine transformation, a root mean square error (RMSE) of 0.58 pixel was obtained in georeferencing the ASTER imagery. For each of the settlements with incidence of BU the percentage prevalence of BU was calculated. Prevalence expresses cases of a disease in terms of the proportion of the population afflicted at a specified time [ 44 ]. It is expressed here as the number of BU cases in a settlement in 1999 divided by the estimated population in 2000 multiplied by 100 to yield a percentage. Spatial factor maps The spatial factor maps generated from the stream sediment geochemistry data for use in the spatial analysis were: (a) map of arsenic-enriched catchment basins; (b) map of arsenic-enriched farmlands; (c) map of arsenic-enriched drainage channels. Arsenic-enriched catchment basins The stream sediment geochemical data for arsenic were initially analysed statistically to determine a threshold value that divides the data into background (normal) classes and anomalous (abnormally high) classes of arsenic concentrations. The data are lognormally distributed and, after removing obvious outliers in the data, a geometric mean of 8.9 ppm As and standard deviation of 2.8 ppm As were obtained. The threshold value was therefore set at 15 ppm As (i.e., approximately the mean plus two standard deviations). The spatial distribution of arsenic was then mapped through the generation of a catchment basin anomaly map in which a sample catchment basin is assigned the geochemical attribute of the corresponding sample [ 45 , 46 ]. Generation of sample catchment basins involved the following steps (using ILWIS): • creation of a raster digital elevation model (DEM) through interpolation of elevation contours; • generation of raster map of drainage lines; and • calculation of sample catchment basin boundaries via an iterative calculation procedure involving the DEM and the raster map of drainage lines. The catchment basin map of arsenic concentrations was then classified into a binary map showing arsenic-normal areas (with ≤ 15 ppm As) and arsenic-enriched areas (with > 15 ppm As) as shown in Figure 2 . About 24% of the study area is occupied by arsenic-enriched catchment basins. Figure 2 Binary map of the catchment basin. Binary map (of the catchment basin) showing arsenic-normal and arsenic-enriched areas. Arsenic-enriched farmlands A supervised classification of ASTER imagery was carried out to distinguish between the major landcover/landuse classes known in the area. These landcover classes are (a) forest areas, (b) residential areas or settlements (bare of vegetation), and (c) farmlands. Using the available landuse/landcover map and topographic map as references, training pixels of known landuse/landcover classes were selected using a colour composite of ASTER bands 2, 3 and 4. These three bands gave the highest optimal index factor (OIF), which indicates the combination of three spectral bands that provide optimum information about landcover [ 47 ]. The box classifier [ 48 ] was chosen for the image classification. The classified image (Figure 3 ), which was also validated in the field, has an overall accuracy of at least 91% with reference to the landcover/landuse map. The classified image indicates that about 91% of the area is farmland. Figure 3 Landcover/landuse map. Landcover/landuse map based on supervised classification of ASTER data. To determine arsenic-enriched farmlands, a Boolean AND operation was performed by crossing the catchment basin anomaly map and the classified landcover/landuse image. About 21% of the total area of farmlands in the classified image is arsenic-enriched. Arsenic-enriched portions of the drainage systems A Boolean AND operation was performed by crossing the catchment basin anomaly map and the raster map of drainage lines. About 22% of the total length of drainage lines is indicated to be arsenic-enriched. Spatial data analysis The inhabitants of a settlement earn their livelihoods by exploiting the resources of the surrounding land. This land influences their exposure to infections and to environmental factors that dispose to infections. Proximity analysis was therefore used to determine spatial relationships between BU prevalence per settlement and (i) arsenic-enriched farmlands and (ii) arsenic-enriched portions of the drainage system. The proximity analysis was carried in two principal steps. First, maps of distances from arsenic-enriched farmlands and arsenic-enriched portions of the drainage system were generated. Second, the point map of BU prevalence per settlement was overlaid on (or crossed with) each of these maps. A buffer is a zone of specified distance around a selected map feature. A GIS creates buffer zones around selected map features such as arsenic-enriched farmlands and arsenic-enriched portions of drainage systems. Around each of these, buffers were set at intervals of 100 m up to 1000 m. Each buffer zone map was crossed with BU prevalence data of settlement to determine how many of these fall within and outside of the buffer. At each increasing interval of 100 m, a test of the significance of the difference of the mean BU prevalence within the buffer and outside of the buffer is made using the t-statistic: where , are the sample means, is the pooled sample variance, n i and n j are the sample sizes from population i and j . Using t ij and degrees of freedom given by n i + n j -2, a t distribution look-up table provides the probability, P that the means are significantly different. Authors' contributions AAD carried out the research and drafted the manuscript. EJMC guided parts of the research and both EJMC and MH reviewed the manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC522823.xml
543577	Thermodynamics of weight loss diets	Background It is commonly held that "a calorie is a calorie", i.e. that diets of equal caloric content will result in identical weight change independent of macronutrient composition, and appeal is frequently made to the laws of thermodynamics. We have previously shown that thermodynamics does not support such a view and that diets of different macronutrient content may be expected to induce different changes in body mass. Low carbohydrate diets in particular have claimed a "metabolic advantage" meaning more weight loss than in isocaloric diets of higher carbohydrate content. In this review, for pedagogic clarity, we reframe the theoretical discussion to directly link thermodynamic inefficiency to weight change. The problem in outline: Is metabolic advantage theoretically possible? If so, what biochemical mechanisms might plausibly explain it? Finally, what experimental evidence exists to determine whether it does or does not occur? Results Reduced thermodynamic efficiency will result in increased weight loss. The laws of thermodynamics are silent on the existence of variable thermodynamic efficiency in metabolic processes. Therefore such variability is permitted and can be related to differences in weight lost. The existence of variable efficiency and metabolic advantage is therefore an empiric question rather than a theoretical one, confirmed by many experimental isocaloric studies, pending a properly performed meta-analysis. Mechanisms are as yet unknown, but plausible mechanisms at the metabolic level are proposed. Conclusions Variable thermodynamic efficiency due to dietary manipulation is permitted by physical laws, is supported by much experimental data, and may be reasonably explained by plausible mechanisms.	Background Carbohydrate restriction as a general strategy for weight loss continues to gain in popularity and its utility and generally protective effect in lipid profile and glycemic control continues to be demonstrated, at least in an experimental setting [ 1 - 4 ]. The subject nonetheless remains controversial. Those critics who grant efficacy of low carbohydrate diets nonetheless contend that they act strictly by caloric restriction and there is no special effect of carbohydrate reduction. Beyond caloric restriction, several studies have shown increased weight loss on low carbohydrate diets compared to isocaloric low fat diets, the so-called metabolic advantage (see table 2 ). Although no clear experimental error has been demonstrated, critics continue to maintain that something must be wrong because the laws of thermodynamics would be violated [ 5 ], that "a calorie is a calorie" [ 6 ] We have previously shown [ 2 , 7 ] that this is not correct and it is our intention here to review the fundamental physics underlying the phenomenon of metabolic advantage. An outline may be described: Can metabolic advantage happen? If so, what mechanisms might account for such a phenomenon? Does it, in fact, occur? Table 2 Isocaloric low carbohydrate (CHO) vs. higher carbohydrate investigations Reference %CHO %CHO Wt. loss(kg) ± SEM p Low High Low CHO arm (no. subjects) High CHO arm Rabast et al (1978) [31] 10 68 14.0 ± 1.4 (25) 9.8 ± 1.0 (20) 0.10 Rabast et al (1981) [32] 12 70 12.5 ± 0.9 (7) 9.5 ± 0.7 (7) <0.01 Golay, Allaz et al (1996) [33] 15 45 8.9 ± 0.6 (22) 7.5 ± 0.5 (21) 0.1 Golay, Eigenheer et al (1996) [34] 25 45 10.2 ± 0.7 (31) 8.6 ± 0.8 (37) 0.13 Piatti et al (1994) [35] 35 60 4.5 ± 0.4 (10) 6.4 ± 0.9 (15) 0.3 Layman et al (2003) [36] 44 59 7.5 ± 1.4 (12) 7.0 ± 1.4 (12) 0.8 Baba et al (1999) [38] 25 68 8.3 ± 0.7 (7) 6.0 ± 0.6 (6) <0.05 Lean et al (1997) [37] 35 58 6.8 ± 0.8 (40) 5.6 ± 0.8 (42) 0.1 Young et al (1971) [39] 7 23 16.2 ± 0.9 (3) 11.9 ± 0.8 (3) <0.05 Greene et al (2003) [40] 5 55 10.4 ± 2.1 (21) 7.7 ± 1.1 (21) 0.25 Metabolic advantage: can it happen? We have previously presented arguments that there is no violation of physical principles [ 2 , 7 ] and, ironically, that suggesting a change in body mass to be independent of macronutrient composition would itself be a violation of the second law of thermodynamics [ 7 ]. Here, we reframe these arguments in a more pedagogically direct way and we provide simple examples. The misunderstanding that continues to be repeated in the expression "a calorie is a calorie" appears to be exclusive reference to the first law of thermodynamics. The difficulty with this theoretical approach is that it is only part of the relevant physics and its relationship to biologic systems. The first law says that in any transformation the total energy in the system can be accounted for by the heat added to the system, the work done by the system on its environment and the change in energy content of all the components of the system. It is important to understand, however, that the first law does not say what the relative distribution between these effects will be for any process. In fact, the first law does not even allow us to say whether the process will occur at all. To understand the progress of a physical change it is necessary to understand the second law which introduces an entity known as the entropy, S, a measure of disorder in all processes. In all real (irreversible) processes, entropy increases, usually written ΔS > 0. The most common marker of increasing entropy is heat, although it is by no means the only evidence for increased entropy. In systems at constant temperature and pressure (i.e. biologic systems)), the first and second law are combined in the Gibbs Free Energy, ΔG, which represents the maximum useful work that can be performed by the process. The actual process however, in general derives less useful work than permitted by the theoretically available ΔG due to inefficiency in energy capture. A proper accounting of entropy and efficiency must be included if we are to understand energy utilization in biological and biochemical systems. Biological systems and thermodynamics It is also important in the discussion of biological systems to understand that they are open systems, i.e. they take in nutrients and oxygen and excrete carbon dioxide, water, urea and other waste products, as well as heat. The importance with respect to weight considerations is that mass and energy are conserved (the more general statement of the first law of thermodynamics), but they are not conserved entirely within the organism. To illustrate the proper interpretation of the first law of thermodynamics consider a subject whose resting energy expenditure is met by the production of 95 moles of ATP. Since oxidation of a single mole of glucose provides 38 moles of ATP, 2.5 moles of glucose will be needed to meet this individual's resting energy requirements. It is important to note that the resultant carbon dioxide, water, and heat are not retained within the organism. The useful retained energy is in the 95 moles of ATP (Figure 1B ). (Similar equations could be written for lipid or protein but we restrict our discussion to glucose for simplicity). Figure 1 A: Oxidation of glucose in a calorimeter is completely inefficient. The products of oxidation are carbon dioxide and water, and all of the energy produced is released as heat. 1B: To illustrate the proper interpretation of the first law of thermodynamics in living organisms we must consider that conservation of matter and energy includes excretion of products into the external environment. None of the products of oxidation (CO 2 and H 2 O) remain within the organism. There is stoichiometric balance and no net weight change. Only the ATP, representing the useful energy, is retained. The wasted heat constitutes 60% of the energy of oxidation, while the efficiency is reflected in the retained ATP, available for reactions in the organism. Body fat stores are signified as TAG (triacylglycerol) 1C . A common way of thinking of weight loss is from reduction of caloric intake. If our subject ingests 2.3 moles of glucose (or equivalent lipid and/or protein) and produces only 90 moles of ATP, then homeostasis will enlist body stores of fat (and/or lean body mass) to yield the additionally required 5 moles ATP. The additional resultant CO 2 and H 2 O (and heat) will be excreted (and radiated) leading to weight loss. 1D : If efficiency is reduced then our subject would have to eat more (e.g. 2.9 moles of glucose, or equivalent lipid/protein) to produce 95 moles of ATP and remain at the same weight. The additional CO 2 and H 2 O produced will be excreted maintaining constant weight. 1E : Under conditions of reduced metabolic efficiency (from 40% to about 38% in this example), 90 moles of ATP will be produced from oxidation of 2.5 moles glucose (or equivalent lipid/protein). The remaining 5 moles ATP needed for homeostasis must be made up from oxidation of body stores of lipid or lean mass. This results in weight loss, exactly as it does for the example of reduced caloric intake (Figure 1C). The illustration above can be compared to the oxidation of glucose in a calorimeter in which no useful energy is obtained and the total energy of oxidation is measured as the heat produced. This process is completely inefficient. A traditional (Atwater) value for glucose obtained in the calorimeter is approximately 4 kilocalories of energy per gram (Figure 1A ). By contrast, the living organism above metabolizes and oxidizes glucose so that approximately forty percent of the energy of oxidation is retained as useful ATP (38 moles per mole of glucose)) whereas sixty percent is released as heat, the inefficiency in this mode of oxidation. The entropy (i.e. the second law of thermodynamics) shows up in this inefficiency . The calorimeter heat can no longer be interpreted in a simple way. The energy stored in useful ATP represents the efficiency of 40% (neglecting the difference in entropy between the structures of the products and reactants). This value approximates the efficiency for oxidation of carbohydrate as well as lipid, whereas proteins are generally oxidized at a lower value of approximately 30–35% (Figure 1B ). Summary of thermodynamics in living organism 1. The second law of thermodynamics dictates that there is an inevitable metabolic inefficiency in all biological and biochemical processes with heat and high entropy molecules (carbon dioxide, water, urea) as the most common products. 2. The first law of thermodynamics is satisfied in living (open) systems by properly accounting for the mass excreted and the heat radiated and exported in high entropy molecules. Weight loss due to reduced caloric intake The most common example of weight loss is reduction of caloric intake. At the risk of oversimplification, if our subject ingests fewer than 2.5 moles of glucose and produces, for example, only 90 moles of ATP from food, then homeostasis would require enlisting endogenous body stores for further oxidation. This oxidation would then provide the additional 5 moles of ATP required. Oxidation of body stores (lipid or lean body mass) will result in production of additional carbon dioxide, urea, water and heat. The excretion of these products will result in weight loss. (Figure 1C ). Weight loss due to increased metabolic inefficiency The implication of the first and second laws of thermodynamics is that reduced efficiency has precisely the same result as reduced caloric intake. One conceptually simple means of reducing efficiency involves the process of uncoupling in mitochondria. ATP is produced in a variety of cellular locations. Glycolysis produces a net of two ATP's per molecule of glucose, in the cell cytoplasm. On the other hand, we recall that 36 additional molecules of ATP are produced from glucose as a result of the mitochondrial TCA cycle and electron transport. A critical part of the process involves the development of a hydrogen ion gradient across the mitochondrial membrane. This concentration gradient provides the energy that is converted into ATP as hydrogen ions pass down the gradient through the ATP synthase particle, entirely analogous to the energy in a high-pressure gas in a cylinder with a movable piston. (The expansion of the gas is like diffusion down a gradient: It does work against the piston). In the mitochondrion the energy of moving down the gradient is captured in ATP, the medium of exchange for the performance of work within cells. This capture of energy, referred to as coupling the energy to the formation of ATP, is the essential process permitting work to be done by living systems. There are known endogenous and pharmacologic agents, which result in uncoupling the formation of ATP from the dissipation of the gradient. Uncouplers such as 2, 4-dinitrophenol bypass ATP synthase and cause hydrogen ion gradient dissipation without ATP formation that can result in organ dysfunction causing death. More modest degrees of uncoupling may be caused by the class of endogenous compounds we know as uncoupling proteins (UCP's). Three different isoforms, UCP1, UCP2 and UCP3 have been identified thus far in mammalian tissues. While the overall and relative physiologic importance of these proteins remains incompletely understood in human tissues, UCP1 has been shown in mice [ 8 ] to result in modest degrees of uncoupling in brown fat. Elevation of fatty acid concentration has been associated with induction of UCP3 and even with pathologic reductions of myocardial efficiency in rat heart [ 9 ]. For purposes of illustration, then, we may consider that there may be physiologic triggers that result in oxidative uncoupling, reducing the overall efficiency of glucose metabolism. For example if efficiency is reduced from 40% to 35%, the result will be the production of only 34 moles of ATP instead of the usual 38. While this represents a mechanism better demonstrated in rats than humans, our subject would require more glucose to make 95 moles of ATP. Now 2.9 moles of glucose would be required to produce 95 moles ATP. Our subject would either eat more and stay at the same weight (Figure 1D ) or would eat 2.5 moles of glucose, the same amount as previously, but would produce less ATP. By eating only 2.5 moles of glucose our subject's metabolism would enlist oxidation of body stores to make up the additional ATP needed for homeostasis. This would result in weight loss exactly as it did for reduced caloric intake. (Figure 1D ). The essence of the second law of thermodynamics is that it guarantees inefficiency in all metabolic processes . However, variation of efficiency is not excluded. In fact, the laws of thermodynamics are silent on the existence of variable efficiency. If efficiency can vary (as in the example of oxidative uncoupling) then "a calorie is a calorie" is no longer a true statement. The role of uncoupling proteins in humans, as indicated, is as yet incompletely defined [ 10 ]. However, thermodynamic principles permit variable efficiency, and its existence must be determined empirically. Metabolic advantage: how could it happen? It is possible that metabolic efficiency may be decreased by oxidative uncoupling as described above. Polymorphisms connecting uncoupling proteins with obesity or propensity to gain weight have been identified in humans [ 11 , 12 ] although these are not firmly established and the effect of dietary intervention is unknown. Other mechanisms are better understood and are described below. Substrate cycling and protein turnover Substrate or "futile" cycles refer to the dynamic process that must accompany the thermodynamic steady state [ 13 ]. In particular, increased cycling of metabolic intermediates utilizes ATP and generates heat. The simplest examples are the numerous kinase-phosphatase pairs that regulate metabolism. In addition, although not generally considered in the category of substrate cycling, inefficiency results from the repeated breakdown and re-synthesis of proteins, lipids, and carbohydrates in cycles that use ATP for no apparent net gain. Such mechanisms, however, far from futile, allow for precision in the regulation of metabolism and constitute one of the uses of ATP. Protein turnover, in particular, provides for error correction or removal of "old" or damaged proteins. The effect of metabolic path on the energetics of oxidation is illustrated in Table 1 which summarizes the analysis from our earlier paper [ 2 ]. In this example, a mole of glucose directly oxidized to CO 2 and water generates 38 moles of ATP with an overall efficiency of about 38.5%. On the other hand, if glucose is first incorporated into glycogen, followed by hydrolysis of the glucose and subsequent oxidation, 2 moles of ATP are lost per mole in this cycle with overall efficiency reduced to 35%. Similarly an amino acid from an "average" protein, when directly oxidized to CO 2 , produces ATP with an efficiency of about 33%. If the amino acid is first incorporated into a protein and later hydrolyzed and oxidized, four ATP's per molecule are used for synthesis of the peptide bond. This reduces the efficiency to 27%. Smaller degrees of inefficiency are seen for lipid cycles (Table 1 ) but multiple cycles may have a cumulative effect. It is estimated, for example, that half of depot fatty acids in triacylglycerol have been through at least one cycle [ 14 ]. It should be apparent that variation in efficiency is not a thermodynamic issue but an empiric question to be determined by the requirements of metabolism. Table 1 Effect of Path on energetics of oxidation Macronutrient & path Mass ATP/mole Kcal/gm Efficiency (%) Glucose → CO 2 180 38 1.54 38.5 Glucose → glycogen → glucose → CO 2 180 36 1.40 35 "Average" AA → CO 2 1.32 33 AA → Protein → AA → CO 2 -4 1.08 27 Palmitate → CO 2 256 129 3.68 40.9 Palmitate → Ketone → CO 2 256 121 3.45 38.3 *Adapted from Feinman, Fine: 2003 Metabolic Syndrome and Related Disorders (1): 209–219 [2] Thyrotoxicosis Thyroid hormone decreases efficiency possibly by mechanisms involving both uncoupling and cycling described above: oxidative uncoupling as well as increased futile cycling of intermediates [ 15 ]. It is observed in thyrotoxic mice that UCP1 decreases efficiency in brown fat at the mitochondrial level [ 8 ]. In humans, the role of UCP1 in thyrotoxicosis is less certain due to the relative paucity of brown fat. On the other hand, activation of the adrenergic system via phosphoenolpyruvate carboxykinase ultimately increases "futile" metabolic cycling of intermediates ([ 15 ]). Thyrotoxicosis is well known to result in weight loss, often with increased food intake and increased generation of heat, indicative of metabolic inefficiency. The use of thyroid hormone has even been suggested therapeutically to induce weight loss in obese individuals, although its toxicity has limited this application. Inefficiency in metabolic processes with weight loss and increased heat generation, therefore, is known to occur on clinical grounds. Even without a complete understanding of the relative importance of different underlying cellular mechanisms in humans, the potential for biochemical processes to reduce their efficiency must be considered established as a feature of mammalian metabolism . Protein induced protein turnover There is abundant evidence that dietary protein stimulates protein breakdown and re-synthesis. In particular, branched chain amino acids, and especially leucine, are documented to act as nutritional signals acting via both the insulin and mTOR signaling pathways [ 16 - 18 ]. On the macroscopic level, the energetic cost of protein turnover is demonstrable as excess heat generated during a high protein meal. Thermogenesis (thermogenic effect of feeding; old name: specific dynamic action) has been defined as the extra heat generated during a meal due to digestion or metabolism. Johnston et al [ 19 ] compared the energy expended during 9 hour intravenous feedings of a high protein meal, vs. an isocaloric high carbohydrate meal; both contrasted with a 9 hour fast. The protein meal, with 70% of its caloric value due to protein, had significantly greater thermogenesis than the high carbohydrate meal (70% of calories from carbohydrate). These data have been reproduced in numerous studies [ 19 - 22 ]. The overall energy costs of protein turnover and synthesis have been estimated in various animal species, including man, and tabulated by Vernon Young ([ 23 ]), based on data from other investigators [ 24 - 26 ]. Despite the substantial experimental difficulties involved, the cost of protein synthesis clusters at around 4–5 kcal/gram in 8 species of birds, marsupials and mammals, including man. The high energetic cost is understandable in view of the multiple ATP-requiring processes involved. The cost of protein turnover can reduce efficiency from 33% to 27%, merely in the formation and hydrolysis of a single peptide bond (requiring 4 ATP's per bond formed: Table 1 ). In addition, protein processes that are ATP-dependent include formation of the ribosomal initiation complex, translation and folding of the protein, and protein degradation (both ubiquitin-dependent and -independent pathways) [ 23 ]. The energy costs of protein turnover could therefore account for a metabolic advantage in high protein diets, independent of carbohydrate content. This mechanism may also contribute to inefficiency in low carbohydrate diets, often high in protein. Gluconeogenesis-stimulated protein turnover in carbohydrate restriction The following hypothesis is suggested from classic studies of starvation done in chronically fasted obese individuals [ 27 , 28 ]. The brain's metabolism requires 100 grams of glucose per day. In the early phase of starvation, glycogen stores are rapidly reduced, so the requirement for glucose, is met by gluconeogenesis. Approximately 15–20 grams are available from glycerol production due to lipolysis, but fatty acid oxidation generally cannot be used to produce glucose. Therefore, protein breakdown must supply the rest of substrate for conversion to glucose in the early phases of starvation. By 6 weeks of starvation, ketone bodies plus glycerol can replace 85% of the brain's metabolic needs, the remainder still arising from gluconeogenesis due to protein. It should be mentioned that, since the fundamental role of ketones is to spare protein, it might be expected that the reliance on protein would actually decrease with time, perhaps relating to the anecdotal observation of "hitting the wall" on weight loss diets. Very low carbohydrate diets, in their early phases, also must supply substantial glucose to the brain from gluconeogenesis. For example, the early phase of the popular Atkins or Protein Power diet restricts dieters to about 20–30 grams of carbohydrate per day, leaving 60–65 grams to be made up from protein-originated gluconeogenesis. One hundred grams of an "average" protein can supply about 57 grams of glucose so 110 grams protein would be needed to provide 60–65 grams glucose. Increased gluconeogenesis has been directly confirmed using tracer studies on day 11 of a very low carbohydrate diet (approx 8 grams/day) [ 29 ]. If indeed, 110 grams of endogenous protein is broken down for gluconeogenesis and re-synthesized, the energy cost, at 4–5 kcal/gram could amount to as much as 400–600 kcal/day. This is a sizable metabolic advantage. Of course, the source of protein for gluconeogenesis may be dietary rather than endogenous. Whereas endogenous protein breakdown is likely to evoke energetically costly re-synthesis in an organism in homeostasis, dietary protein may conserve energy. The source of protein for the observed gluconeogenesis [ 29 ] remains an open question, but there is no a priori reason to exclude endogenous rather than dietary sources. This is therefore a hypothesis that would need to be tested. The extent to which the protein for gluconeogenesis is supplied by endogenous protein would explain very high-energy costs. It should be noted, however, that even if limited to breakdown of dietary protein sources, there would be some energy cost associated with gluconeogenesis. Metabolic advantage: does it happen? Having established that there is no theoretical barrier to metabolic advantage and that there are plausible mechanisms that could account for such an effect, we must ask whether it can be demonstrated experimentally, that is, whether the proposed effects are of sufficient magnitude to be a practical feature of weight reduction strategies, in particular very low carbohydrate diets. If so there will be increased weight loss for the same caloric intake, or metabolic advantage. A recent animal model provides support for greater metabolic inefficiency in rats fed carbohydrate restricted diets compared with higher carbohydrate, leading to excess weight loss [ 30 ]. Human data in Table 2 illustrates 10 clinical trials of isocaloric diets with a lower versus higher carbohydrate arm in each trial [ 31 - 40 ]. It can be seen that the lower carbohydrate arm in 9 of 10 studies demonstrates increased weight reduction in comparison with the higher carbohydrate arm. Three of the studies show statistical significance (p < 0.05 or better). Even without statistical significance of individual studies, however, the likelihood that the lower carbohydrate arm would have an advantage in 9 of 10 studies is equivalent to the likelihood of 9 coin toss experiments having excess heads in comparison to excess tails. The 9 th binomial coefficient shows this probability to be p < 0.01. While the above suggests the possibility of metabolic advantage, it does not prove it, nor do we know the magnitude of the effect, or the factors that control it. The studies above were chosen from among those quoted by many of the authors who have disputed the existence of metabolic advantage. Nonetheless, a formal meta-analysis would be necessary to avoid the possibility of conscious or unconscious bias in their selection. Further, it would be necessary to establish evidence that energetically costly metabolic processes are more prevalent in low carbohydrate diets than in diets of higher carbohydrate content. Whereas the proposed mechanisms are plausible, they need to be proven. Conclusions Thermodynamics is not the limiting factor behind the concept of metabolic advantage. On the contrary, thermodynamics guarantees inefficiency in all metabolic processes and is silent on the possibility that inefficiency may be augmented in some instances. A familiar example of inefficiency is thyrotoxicosis, with attendant weight loss and heat generation despite unchanged or increased caloric consumption. The theoretical possibility of inefficiency and metabolic advantage due to macronutrient compositional change exists, but demonstration of the phenomenon can only be resolved experimentally. Isocaloric dietary studies with a low vs. a higher carbohydrate arm support the experimental possibility of metabolic advantage. A formal meta-analysis would be required to evaluate this more objectively. Further studies, including tracer methods, would be required to establish mechanisms. The presence of high quantities of dietary protein (often a feature of low carbohydrate diets) is known to stimulate protein turnover, an energetically costly process. However, it is unclear whether this is the only factor, or whether it is necessary for metabolic advantage to occur. In particular, obligate gluconeogenesis from endogenous sources may also contribute to induction of protein turnover. Competing interests The author(s) declare that they have no competing interests.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC543577.xml
554779	General practitioner attitudes to the care of people with epilepsy: an examination of clustering within practices and prediction of patient-rated quality of care	Background There is wide variation in the quality of care provided by primary care practices to individuals with chronic illnesses. Individual doctor attitudes and interest have been demonstrated to influence patient outcomes in some instances. Given the trend towards larger practices and part-time working, continuity of care is likely to fall and thus practice-based rather than individual general practitioner attributes and attitudes are likely to become increasingly important. The aim in this paper was to examine the extent to which individual general practitioner (G.P.) attitudes to the care of people with epilepsy cluster within practices and predict patient-rated quality of care. Methods The sample consisted of 1255 people with active epilepsy (a recent seizure or on anti-convulsant medication for epilepsy) and 199 GPs from 82 general practices. Measures of GP attitudes (a 17-item GP attitudes questionnaire) and patient-rated quality of epilepsy care were obtained. 1210 individuals completed initial questionnaires and 975 patients filled in final questionnaires one year later. Responses were achieved from 64 practices (83% of total) and 115 GPs (60% of total). Results 2 main factors were found to underlie GP attitudes to the care of people with epilepsy and these demonstrated clustering within practices "epilepsy viewed as a primary care responsibility" (Eigenvalue 3.98, intra-class correlation coefficient (ICC) 0.40), and "medication skills"(Eigenvalue 2.74, ICC 0.35). GP-rated scores on "epilepsy care being a primary care responsibility" were a significant predictor of patient-rated quality of GP care (p = 0.031). Other contributory factors were seizure frequency (p = 0.044), and patient-rated "shared decision making" (p = 0.022). Conclusion Specific general practitioner attitudes to the care of people with epilepsy cluster within practices and are significantly associated with patient-rated quality of epilepsy care. It is important to take these findings into consideration when planning primary care interventions to ensure people with epilepsy receive the benefits of available medical and surgical expertise.	Background There is wide variation in the quality of care provided by primary care practices to individuals with chronic illnesses [ 1 ]. Individual doctor attitudes and interest have been demonstrated to influence patient outcomes in some instances [ 2 , 3 ] and it has been argued that specific attitudes are more predictive of behaviour than general attitudes[ 4 ]. For chronic diseases, in line with the distinction proposed by Katz [ 5 ], these doctor attitudes may be separated into perceptions of knowledge, skills and personal preferences. The importance of these specific doctor attitudes on patient outcomes is however largely unknown. There is a marked trend to larger partnerships in primary care practices and more flexible working practices. It is likely, therefore, that continuity of care will continue to fall, and that patient experience of care of a particular condition will be based on contact with more than one general practitioner [ 6 ]. Thus practice-based rather than individual general practitioner (GP) attributes and attitudes are likely to become increasingly important. The extent to which GP attitudes to specific chronic conditions cluster within practices, is however currently unknown. There is evidence that where attitudes within a group are shared this enhances the influence of individual attitudes on behaviour [ 7 ]. Thus, on the basis of this observation, if attitudes are shared by general practitioners within practices, these group-based attitudes are then more likely to influence GP behaviour and the quality of care provided by the clinician. In the next few years there is likely to be considerable reorganisation of the way in which epilepsy care within primary care is delivered, with GPs taking on a more active role in providing care. Information on how individual general practitioners view and value their role in providing epilepsy care is considered as important [ 8 ]. However, what may be more important is whether or not these views are shared within the practice and if these attitudes influence the quality of care provided by the practice. If this is the case, then taking GP attitudes in a given practice into account will be crucial in deciding how primary care services for people with epilepsy are best organised and improved. In this paper, data from a completed community-based study on people with epilepsy are used to examine the following questions: 1. To what extent do individual general practitioner attitudes to the care of people with epilepsy cluster within practices? 2. Do general practitioner attitudes predict how people with epilepsy rate the quality of the general practitioner care of their epilepsy? Methods General practitioners and adults with epilepsy taking part in an intervention study in Greater Manchester provided information for this study. The results of the intervention study (a prompt and reminder card for general practitioners to complete, held by patients or placed in their medical records and used opportunistically over the course of a year) have been reported [ 9 ]. No group differences in patient rated outcome measures were found for the intervention study [ 9 ]. Ethical approval was obtained in the 4 areas of Greater Manchester from where patients were approached. The patients who consented to participate in the study had their medical records examined to extract data on recording of clinical information about epilepsy and other markers of quality of care. Patients were also sent questionnaires for self-completion. These included both a generic quality of life measure- the EUROQOL 5D[ 10 ] and a disease specific quality of life and quality of care measure, the "Living with Epilepsy" questionnaire which has been psychometrically tested and shows good reliability and validity [ 11 ]. Self-rated seizure frequency (included in the "Living with Epilepsy" questionnaire) was based on the response to a question "How many epileptic attacks have you had in the past year" with the 3 response categories being "None", " Less than one a month" and "One or more a month". General practitioners completed a 17-item GP epilepsy attitudes questionnaire at the end of the study. Responses to items such as "I feel comfortable changing the type of anti epileptic drugs in my patients" were scored using a Likert scale. The attitudes scale was developed and validated for a previous study and results reported in an earlier paper [ 12 ]. Statistical analysis Mean practice scores for each item on the GP attitudes questionnaire were computed and significant factors underlying the grouping identified (using eigenvalues >1.2 and the Scree test.). Clustering of responses on the attitudes questionnaire were examined using the intra-class correlation coefficient both for individual items as well as for a mean score of responses to items loading on each of the main factors. Finally linear regression analysis with the patient-rated quality of care provided by the practice as the dependent variable and GP attitudes and other patient derived measures (such as seizure frequency, age, gender, other long term illness) as independent variables was carried out (using aggregated GP attitude and patient scores). Intervention group was adjusted for. P values of 0.05 and 95% confidence intervals were used to assess significance. Results 1255 patients consented to participate in the study and 975 patients filled in final questionnaires. 199 GPs from 82 practices consented to participate in the study. Responses were obtained from 115 GPs (60% of total) from 64 practices (83% of total). 29 practices had a single respondent. These practices were excluded from the analysis of attitude clustering. In this study 54% of individuals were seizure-free in the previous year ("controlled" seizures) and 46% had reported a seizure in the previous year ("uncontrolled" seizures). Factor analysis and clustering of GP attitudes Factor analysis with varimax rotation was undertaken on aggregate GP responses. Four factors had an eigenvalue of above 1.2. Three of these factors were selected after the scree test. Both the Kaiser-Meyer-Olkin measure of Sampling adequacy test (0.778) and Bartletts test for sphericity (Chi-square 413.7, Df 120, p < 0.0001) suggested that factor analysis was appropriate for this data set. Using guidelines for identifying significant factor loadings based on sample size from Hair et al.[ 13 ] a cut-off of 0.65 was used. Individual items within each factor were used to generate mean factor scores. These mean factor scores were normally distributed. Responses to the 11 questions that were included in the first three factors were further examined. The aim was to detect if significant clustering of responses to these items occurred within practices. The average cluster size was 2.74. The results of the factor analysis and clustering of attitudes for the two main factors are listed in Table 1 . The other two factors were excluded. The first excluded factor was not clinically meaningful with only two disparate items loading on it ("epilepsy care straightforward", "epilepsy patients viewed as being well-informed"). The second excluded factor explained less than 10% of variance and only had one item loading on it ("self-perceived knowledge of epilepsy"). Table 1 Two main extracted factors, significant factor loadings and intraclass correlation coefficients (ICC) for general practitioner attitudes within practices Factor loading 1 ICC 2 Factor 1: "Primary care responsibility"(Eigenvalue 3.98, 24.9% of variance explained)-mean scores 0.40 "Not too time pressured to take on epilepsy care" 0.785 0.37** "GP has primary responsibility for organising follow up care" 0.769 0.13 "Epilepsy care not too difficult to organise" 0.767 0.19* "Epilepsy care not a specialist responsibility" 0.732 0.34 "Epilepsy care should be based in general practice" 0.684 0.44 "Annual structured review should be carried out in primary care" 0.657 0.10 Factor 2: "Medication skills"(Eigenvalue 2.74, 17.1% of variance explained) 0.35 "Comfortable adjusting dose of medication" 0.724 0.31** "GP responsible for adjusting treatment if more fits" 0.718 0.25* "Comfortable adjusting type of medication" 0.655 0.17 p < 0.05, ** p < 0.01 1 (based on mean GP scores per practice) 2 (based on individual GP scores in practices with >1 respondent) Do GP attitudes predict how patients rate the quality of GP care of their epilepsy? Data from 60 practices where both patient and general practitioner data were available were used in the linear regression analysis. As the data were obtained at the end of an intervention study, the intervention group was also included as an independent variable. The results of this analysis are given in Table 2 . Significant predictors of patient-rated quality of GP care were patient seizure frequency and patient -rated "shared decision making" and GP -rated score on "epilepsy care being a primary care responsibility" (Factor 1). Recording of clinical information about epilepsy was not a significant predictor of patient-rated quality of GP care. Table 2 Linear regression analysis: GP and patient predictors of patient rated satisfaction with GP care of epilepsy Unstandardized Coefficient Standardized Coefficient t value Sig. 95% Confidence intervals for B B Std Error Beta Lower bound Upper bound Patient measures Age -.012 .006 -.19 -1.80 .079 -.025 .001 Gender -.4 .258 -.19 -1.55 .128 -.921 .120 Long term health problems other than epilepsy -.141 .255 -.05 -.55 .583 -.655 .373 Anxiety score .051 .028 .24 1.80 .078 -.006 .108 Depression scores -.048 .041 -.17 -1.16 .255 -.131 .036 Ease of talking to GP about epilepsy .392 .348 .13 1.128 .266 -.309 1.109 GP takes views of epilepsy into account ("shared decision making")* .931 .390 .31 2.38 .022 .144 1.719 Seizure frequency* .306 .148 .26 2.07 .044 .008 .604 GP measures Epilepsy as primary care responsibility* (factor 1) .154 .069 .29 2.23 .031 .015 .294 Medication skills (factor 2) -.033 .059 -.07 -.564 .576 -.152 .086 Data adjusted for intervention group r 2 = 0.635, adjusted r 2 = 0.525, standard error = 0.224 * p value <0.05 Some further bivariate analyses were also undertaken. Recording of clinical information about epilepsy by GPs was not significantly associated with the GP-rated score on epilepsy care being a primary care responsibility but was associated with seizure frequency. Discussion In this study two main factors ("epilepsy viewed as a primary care responsibility" and "medication skills") were found to underlie GP attitudes to the care of people with epilepsy. Responses to questions constituting these factors demonstrated a high and significant level of clustering within practices. The main factor that accounted for the largest proportion of variance, general practitioner-rated "epilepsy viewed as a primary care responsibility", significantly predicted patient-rated quality of care. Patient-rated shared decision-making and seizure frequency were other significant predictors of patient-rated quality of GP epilepsy care. Recording of clinical information by GPs about epilepsy was not associated with GP attitudes to epilepsy care but was related to patient seizure frequency. In this study general practitioner attitudes to the care of people with epilepsy were found to cluster within practices to a considerable extent. This has not previously been shown in the U.K. A recent Dutch study [ 14 ] showed that GPs working in the same partnership showed more resemblance in overall attitudes to patient care and behaviour than GPs not working in the same partnership and hypothesised that social processes in partnerships and local circumstances may be particularly relevant. The present study has quantified these intra-practice GP similarities in terms of attitudes to one specific chronic condition. Moreover the results of this study also demonstrate that certain general practitioner attitudes predict patient-rated quality of care provided by the practice. The high level of clustering of GP attitudes and the effect of these attitudes on patient-rated outcomes in terms of quality of care, may have important implications in determining the effectiveness of practice level interventions in primary care. These results suggest, firstly, that when planning educational interventions, changing GP attitudes within practices should also be a key aim and, secondly, to focus on changing attitudes for the practice as a whole rather than simply for individual general practitioners. In addition, for practice level intervention studies (especially those using patient-rated quality of care as an outcome measure) an estimate of clustering of doctor attitudes as well as estimates of clustering of patient responses when carrying out power calculations should be incorporated to avoid making a Type 1 error. There is relatively little information of the relationship of GP attitudes to patient ratings of the quality of GP epilepsy care. Existing evidence suggests that GPs with a special interest in a particular condition improves outcomes [ 2 ]. The results of the present study extend these findings by highlighting the importance of specific attitudes (accepting a key role in management) rather than perceptions of specific skills (skills in medication management) in predicting patient rated quality of care. The results of a multilevel analysis examining patient and doctor predictors of patient satisfaction from the Netherlands [ 15 ] suggested that most of the variance in "patient satisfaction" scores was at the patient level (age, morbidity and previous negative experience with the GP being the main predictors) with only 5–10% of the variance in "patient satisfaction" being at the doctor level. However in that Dutch study [ 15 ], specific GP attitudes were not included as predictors and the "patient satisfaction" score was a composite score incorporating measures of accessibility, availability, humaneness of the GP and information provision. Patient-centred communication skills are known to be associated with improved patient satisfaction [ 16 ] and our analysis indeed found that patient-rated shared decision making ("GP took my views into account") was another significant predictor. Patient ratings of the quality of care do vary according to whether individuals have controlled or uncontrolled seizures. Individuals with controlled seizures rate the quality of care provided higher than individuals with uncontrolled seizures. However why the ratings of care provided are higher is not clear as individuals with controlled and uncontrolled epilepsy differ from each other in other characteristics that may influence quality ratings apart from seizure frequency (e.g. depression scores, social functioning). At practice level GP attitudes are not related to mean practice patient seizure frequency. Although it is likely that individual GP attitudes will be influenced by whether an individual patient has controlled or uncontrolled seizures it is not possible to empirically demonstrate this relationship, as nearly all general practitioners will see a mix of individuals with "controlled" and "uncontrolled" seizures in a given year. Their attitudes to the care of people with epilepsy will be influenced by this spectrum of epilepsy severity (and often to a greater extent by other factors including significant events with individual patients). In terms of limitations of the results, some practices did not consent to take part in this intervention study and not all GPs who participated completed questionnaires. However there were no significant differences between practices that participated and did not participate in terms of size, average deprivation or training status [ 9 ]. Moreover responses to the GP questionnaire were received from over 80% of practices that participated and 60% of the doctors that participated. Aggregate scores were used when doctor and patient views were analysed. This will reduce variability and may result in a loss of statistical power. However given that many different patient-doctor encounters are likely within a given year this approach was the most pragmatic. Although the results were obtained at the end of an intervention study that may have influenced attitudes one of the groups in the study was a control group and no significant differences in GP attitudes between groups was found. Furthermore, results on GP attitudes in the present study were very similar to those found in a previous survey using this scale [ 12 ]. Conclusion Specific general practitioner attitudes to the care of people with epilepsy are significantly associated with patient-rated quality of epilepsy care and cluster within practices. It is important to take these findings into consideration when planning interventions and services. General practitioners need to have good knowledge and skills in the management of epilepsy and should be aware of and utilise current guidelines for good clinical epilepsy care [ 17 - 19 ] to fully utilize medical and surgical expertise in managing epilepsy. Recognising and addressing general practitioner attitudes to the care of people with epilepsy may be important in ensuring these goals of good epilepsy care are met. Competing interests The author(s) declare that they have no competing interests. Authors' contributions AT designed and ran the study, undertook the analysis and wrote the manuscript. MR was involved in the design and running of the study and edited the paper. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC554779.xml
551615	Familial hypercholesterolemia in St.-Petersburg: the known and novel mutations found in the low density lipoprotein receptor gene in Russia	Background Familial hypercholesterolemia is a human monogenic disease caused by population-specific mutations in the low density lipoprotein (LDL) receptor gene. Despite thirteen different mutations of the LDL receptor gene were reported from Russia prior to 2003, the whole spectrum of disease-causing gene alterations in this country is poorly known and requires further investigation provided by the current study. Methods Forty-five patients with clinical diagnosis of FH were tested for the apolipoprotein B (apoB) mutation R3500Q by restriction fragment length analysis. After exclusion of R3500Q mutation high-sensitive fluorescent single-strand conformation polymorphism (SSCP) analysis and automatic DNA sequencing were used to search for mutations in the LDL receptor gene. Results We found twenty one rare sequence variations of the LDL receptor gene. Nineteen were probably pathogenic mutations, and two (P518P, T705I) were considered as neutral ones. Among the mutations likely to be pathogenic, eight were novel (c.670-671insG, C249X, c.936-940del5, c.1291-1331del41, W422X, c.1855-1856insA, D601N, C646S), and eleven (Q12X, IVS3+1G>A, c.651-653del3, E207X, c.925-931del7, C308Y, L380H, c.1302delG, IVS9+1G>A, V776M, V806I) have already been described in other populations. None of the patients had the R3500Q mutation in the apoB gene. Conclusions Nineteen pathogenic mutations in the LDL receptor gene in 23 probands were identified. Two mutations c.925-931del7 and L380H are shared by St.-Petersburg population with neighbouring Finland and several other mutations with Norway, Sweden or Denmark, i.e. countries from the Baltic Sea region. Only four mutations (c.313+1G>A, c.651-653del3, C308Y and W422X) were recurrent as all those were found in two unrelated families. By this study the number of known mutations in the LDL receptor gene in St.-Petersburg area was increased nearly threefold. Analysis of all 34 low density lipoprotein receptor gene mutations found in St.-Petersburg argues against strong founder effect in Russian familial hypercholesterolemia.	Background Familial hypercholesterolemia (FH) (OMIM #143890) is one of the most common monogenic human diseases. It is inherited as an autosomal dominant trait with the prevalence of the heterozygous form conventionally considered to be about 1 of 500 in most populations [ 1 ]. Elevated blood serum cholesterol is due to impaired removal of low-density lipoproteins (LDL) from blood by LDL receptors, and it is associated with early onset coronary artery disease and myocardial infarction. Impairment of LDL receptor function usually results either from the absence or deficiency of the LDL receptor (OMIM 606945) itself or from a common mutation (R3500Q) in the gene of the receptor's ligand, apolipoprotein B (OMIM +107730) causing type B of the autosomal dominant hypercholesterolemia (OMIM #144010) [ 2 ]. Furthermore, a third type of monogenic autosomal dominant hypercholesterolemia (OMIM #603776) is due to the recently discovered defects in the proprotein convertase PCSK9 gene (OMIM 607786) [ 3 ]. A form of recessively inherited hypercholesterolemia (OMIM #603813) has a prevalence of less than 1:10 000000 and is due to defects in the LDL receptor adaptive protein ARH (OMIM *605747) [ 4 , 5 ]. Even though many forms of monogenic hypercholesterolemia are known, only apoB gene and LDL receptor gene variations seem to contribute significantly to the CHD morbidity in most populations [ 6 ]. The spectrum of LDL receptor mutations varies between different human populations and more than 900 mutations in the LDL receptor gene have been characterized worldwide [ 7 - 9 ]. A recent review [ 9 ] shows a clear difference in the LDL receptor gene mutations spectra for Western European countries, but this review gives nearly no data on the genetics of FH in the Eastern European countries and Russia. However, many mutations were described from Poland [ 10 ] and Bulgaria [ 11 ] and already thirteen different LDL receptor gene mutations have been published from the Russian population prior to 2003 [ 12 ]. In the present research we expand the study of the molecular genetic basis for FH in St.-Petersburg known to be the most well studied region of Russia in respect to FH-causing mutations and report 21 mutations, previously unknown in Russia. Methods Patients Patients with FH were recruited from two lipid clinics of St.-Petersburg, namely Institute of Human Brain and Institute of Experimental Medicine. The clinical diagnosis of familial hypercholesterolemia was based on the following criteria: highly elevated plasma total cholesterol and LDL-cholesterol, presence of tendon xanthomata, corneal arcus or both, and positive family history of myocardial infarction and hypercholesterolemia with at least one first-degree relative affected. Forty-five probands fulfilling at least two of three criteria listed above were selected for the study. Full data of patients including their lipid data are given in the Discussion section (see Table 2 ). Informed consent was obtained in each case for DNA testing procedures. Biochemical procedures Genomic DNA was extracted from blood white cells using a standard method [ 13 ]. The patients were initially tested for the apoB mutation R3500Q by restriction fragment length analysis [ 14 ]. Exons of the LDL receptor gene were then amplified [ 15 ], and PCR products were subjected to gel-electrophoresis followed by ethidium bromide or silver staining of DNA to exclude non-specific amplification. Single-strand conformation polymorphism (SSCP) analysis was performed in an ABI 377 DNA Sequencer (PE-Applied Biosystems) sequencing device using 4,25% non-denaturing MDE polyacrylamide gels (Cambrex) at 20°C. The gels were run under two different conditions, i.e. standard MDE gel or with the addition of 5% glycerol to the MDE gel. Automated DNA sequencing was performed in part in the ABI 377 DNA Sequencer (PE-Applied Biosystems), in part in an ALFExpress-2 DNA sequencer (Amersham Life Sciences), using primers for routine PCR DNA amplification. In the case of samples bearing the deletions, PCR products were cloned into a commercially available vector (TACLONE, Medigen, Novosibirsk) and sequenced using universal and reverse primers. Results None of the 45 patients had the apoB R3500Q mutation, whereas 25 patients had mutations in the LDL receptor gene. Large genomic rearrangements in the LDL receptor gene were previously shown to be an uncommon cause of FH in St.-Petersburg [ 16 ] and had been excluded in most patients of the current group. Therefore we restricted our search to point mutations and to minor deletions and insertions. We identified 21 sequence variations (Table 1 ) of which two probably were not pathogenic (in the following considerations we give numbering of aminoacids in the LDL receptor according to Yamamoto's nomenclature, [ 17 ] ). The T705I mutation (known also as FH Paris-9) is not associated with elevated serum cholesterol [ 18 ], and the transition c.1617C>T (P518P) is synonymous. We consider the remaining 19 variations to be pathogenic. Eleven mutations have been described in other populations, but to our knowledge the remaining 8 mutations have not been described so far [ 7 , 8 ]. From two of these nucleotide substitutions (C249X and W422X) the stop codon arises, four are frame-shift mutations leading to premature stop codons (FsK202:S205X; FsK290:N309X; FsV409:S423X; FsV597:A622X), and two result in amino acid substitutions (D601N and C646S). Sequencing of the cloned mutant allele bearing the five-nucleotide deletion (c.936-940del5 or FsE291) is demonstrated on Fig. 1 . Mutations c.313+1G>A, c.651-653del3, C308Y and W422X were found in two probands each. Rapid tests were developed for most of the mutations and all of those were confirmed by using these methods (Table 1 ). Various detection methods, including heteroduplex analysis, restriction enzyme tests and SSCP for several mutations from the list are illustrated by Figures 2 , 3 . Cosegregation of the mutations and elevated blood serum cholesterol was demonstrated in seventeen families (see Fig. 2 , 3 and Table 1 ). Totally the mutations were confirmed in 9 relatives and excluded in 27 members of the proband's families. Most important the diagnosis of FH was excluded in 9 and set in 2 children of probands before adulthood, i.e. prior to age 18. In these children mutation detection was of crucial importance to set the diagnosis and to suggest further life style. In case of the Q12X mutation, the c.97C>T transition leads to occurrence of a new Mae I restriction site (CTAG) in exon 2 that allows a simple detection of this mutation. Restriction enzyme test was performed for the proband and her 4 descendants (Fig. 2 ). The presence of the mutation was confirmed in the son of the proband and excluded in the daughter and two grandchildren. In some other mutations no restriction enzyme tests can be developed for their rapid detection. For example, it is true for the recurrent mutation IVS3+1G>A (c.313+1G>A) that was identified by means of SSCP and verified by sequencing in probands from two unrelated families. SSCP patterns on silver-stained gels differ strikingly in patients with and without the mutation. The presence of the mutation was confirmed in the son of the proband by SSCP. In heterozygotes a deletion of 41 nucleotides at c.1291-1331 in exon 9 (mutation FsV409:S423X) results in occurrence of specific heteroduplexes. Besides, a PCR fragment of smaller size as compared to the normal allele is revealed in silver-stained polyacrylamide gels (Fig. 3 ). This mutation was identified in the son and daughter of the proband and excluded in the second daughter and in her child (Fig. 3 ). Discussion Previously 13 mutations in the LDL receptor gene have been reported in FH patients residing in St.-Petersburg [ 12 ] (see Table 3 ). Eight of those have not been reported in other populations. Our study revealed 21 mutations in the LDL receptor gene, out of which only T705I was previously reported from St.-Petersburg [ 12 ]. Nineteen of those 21 mutations are likely to be pathogenic. We exclude P518P (CCC>CCT; c.1617 C>T) from the list of pathogenic mutations since the transition c.1617 C>T results neither in an amino acid substitution nor in appearance of the new consensus splicing sequences. P518P mutation was found in the proband with mutation C646S, but it was not clarified if the mutations were in cis- or trans- position. Also we consider the T705I variant not to be a primary cause of FH, since the I705 allele itself is not associated with elevated cholesterol level [ 18 ] and the probable hypercholesterolemic effect of this mutation may be due to its linkage with other pathogenic LDL receptor gene mutations. Indeed, a variation in intron 7 (c.1061-8T) of unclear functional significance was shown to be very tightly linked to the I705 allele [ 19 , 20 ]. We have not searched for this intronic variant in the LDL receptor in the patient with the T705I substitution. Among pathogenic mutations reported here, eleven have been described in other populations (Table 1 ) [ 7 , 8 ], and eight are novel. Six of the novel mutations lead to premature stop codons and result in truncated protein chains that probably lose their function. One out of these truncating mutations (c.936-940del5 or FsE291) also changes invariant nucleotides nearby exon-intron junction and thus may affect splicing. The D601N missense mutation, causing substitution of aspartic acid by asparagine has not been reported before. It seems likely that such a substitution might cause the loss of function since one other mutation in the same codon (D601Y) was described in familial hypercholesterolemia subjects [ 7 ]. The C646S also has not been reported before, but 5 other mutations affecting this codon have been described, one of which, C646Y (FH French Canadian-2), results in a transport defective protein (mutation class 2A) [ 21 ]. We, therefore, find it very likely that the C646S mutation is also pathogenic, but expression studies are needed to justify the effect of both missense mutations D601N and C646S. Previously described mutations were considered to be pathogenic due mostly to their listing in FH mutation databases even though functional studies were not systematically performed. Indeed, V806I mutation (known as variant FH New York -5) [ 22 ] occurs in the LDL receptor internalization signal NPVY, for which the consensus sequence NPxY is given (where X is not a conserved aminoacid) [ 23 ] and thus the substitution of isoleucin for valine may be not crucial for the LDL receptor function. The V776M mutation may have effect on LDL receptor mRNA splicing rather than to be realized on the protein level since the V776M mutation changes the invariant G at the 3' end of exon 16. However, this mutation is likely to be pathogenic since it was reported already in patients from La Habana, Cuba [ 24 ]. The apoB R3500Q mutation was not detected in any of our patients. This finding is in agreement with the previous observation that the R3500Q mutation had not been found in St.-Petersburg [ 25 ]. The mutation has been detected in another part of Russia only in 2 of 71 patients with symptoms of familial hypercholesterolemia [ 26 ]. The apoB R3500Q mutation is almost exclusively found in Caucasian individuals, and almost all subjects with the mutation carry the same haplotype. One of the highest frequencies of the mutation has been found in the Swiss population (approximately 1/200) [ 27 ], and Miserez and Muller [ 27 ] hypothesized that the mutation may have arisen in Switzerland 10,000 – 6,000 years ago. The prevalence of the mutation declines with increasing distance from the Central Europe [ 27 , 28 ], and the prevalence of the apoB R3500Q mutation is therefore expected to be low in the St.-Petersburg area (< 1/1000). A precise estimate would require a random sampling of the general population. Mutations in LDL receptor gene typical for various ethnic groups were revealed in St.-Petersburg (Table 1 ). In particular, several mutations found in Denmark, Finland, Norway and Sweden are present in the St.-Petersburg population. The ethnic origin of these families is unclear and according to questioning and family names no evidence of a Scandinavian origin of probands was obtained. Only exception from our previous FH group was the proband with the E207X mutation who stated his German roots and indeed this mutation was previously reported from several families from Germany [ 29 ]. Interestingly, no specific Slavic or Eastern-European founder mutations were found in St.-Petersburg when comparing the LDL receptor mutation spectrum of Russia to those of Poland [ 10 ], Bulgaria [ 11 ] or Czech Republic [ 30 , 31 ]. G571E was found in Russia, Czech Republic and Poland but this mutation was also reported from many other countries worldwide. The C188Y mutation found in Russia [ 32 ] and in Czech Republic [ 30 , 31 ] cannot be considered a founder Slavic mutation, since it was reported from unique families in each country. Only one candidate for a Russian founder mutation is C139G identified in four unrelated families in different regions of the country, but this mutation is absent in related nations of the Eastern Europe as well as in other countries in the world. In our study we were able to find LDL receptor gene pathogenic mutations in 23 of 45 patients. Many methods are considered to be superior to routine isotopic SSCP when screening for mutations in DNA (see e.g. [ 33 ]for comparison of SSCP and DHPLC). However, we believe that mutations could have been overlooked due to intrinsic limitations of SSCP method in our hands only in few cases. Fluorescent SSCP, used in the current study, can be superior even to DHPLC and is a sensitive method for mutation screening, especially when different gel electrophoresis conditions are applied [ 34 ]. Recent studies indicate, that SSCP run under two different conditions detect up to 96% of heterozygous variations (P.H. Nissen, unpublished results). In our study LDL receptor gene mutations were found in 56% (14 out of 26) patients selected by SSCP and in 53% (10 out of 19) patients which DNA was subjected to direct sequencing of all gene exons (Table 2 ). We do not find it likely that many patients are underdiagnosed due to presence of large genomic rearrangements. In a study conducted in St.-Petersburg, in the partly overlapping sample of FH patients, only one case of a large deletion was found in the sample of 50 probands, giving the rough estimate of genomic rearrangements 2% [ 16 ]. In fact in mixed populations (such as that of that of St. Petersburg) the contribution of large rearrangements to the spectrum of pathogenic mutations seems to vary from 6% in Great Britain [ 35 ] to 2.5% in English-speaking Canadians [ 36 ]. More important, some of intronic mutations leading to defects of splicing could be missed, since the design of primers [ 15 ] mostly following recommendations of Hobbs et al. [ 22 ] with the only exclusion for primers to amplify exon 3 allowed analyzing only 14 out of 34 intron-exon boundaries in the LDL receptor gene. Recent investigation [ 37 ] demonstrated that a high percent of previously missed LDL receptor mutations may be localized in introns. In the cited study [ 37 ] in the patient sample after modifying the gene analysis procedure nearly 27% of patients turned out to have intronic mutations, despite the functional significance of these mutations have still to be validated. In the apoB gene we only looked for the R3500Q mutation, and the possibility that other mutations in that gene are responsible for hyperlipidemia cannot be ruled out. This possibility is unlikely, since the apoB gene has been studied extensively in other laboratories and no other pathogenic variants have been determined. To conclude we cannot definitely rule out the possibility that some undiscovered mutations in the LDL receptor gene have not been found. However, the possibility that the patients have mutations in a different gene involved in FH still remains since the mutation in the PCSK9 gene [ 3 ] were not tested in this study. With the two exceptions, mutations previously reported from Russia [ 12 , 38 ] have been confined to a single family. The exceptions were the mutations G197del (FH-Lithuania), found in high percent of Ashkenazi Jewish families with FH from St.-Petersburg [ 39 ], and the C139G mutation found in two Slavic families from St.-Petersburg [ 40 , 12 ], one family in Novosibirsk [ 38 ] and one family in Moscow [ 41 ]. In this study 4 mutations, namely c.313+1G>A, c.651-653del3, C308Y and W422X were found in two unrelated families. Together with the previous findings we discovered 34 mutations in the LDL receptor gene in St. Petersburg of which only six were detected in more than one family. Fourteen LDL receptor gene mutations were discovered by the other Russian FH team from Moscow, Russia out of which only C139G is shared with St.-Petersburg population [ 12 , 41 ]. To date a total of 47 different FH mutations are known from Russia. The data presented here enlarge the spectrum of mutations found in the Russian population and make us regard it as genetically heterogeneous. Conclusions We identified nineteen pathogenic mutations in the LDL receptor gene in 23 probands and two probably neutral mutations. In our study only four mutations in the LDL receptor gene (c.313+1G>A, c.651-653del3, C308Y and W422X) were found to be recurrent, i.e. all of those were found in two apparently unrelated families. Together with the data obtained earlier [ 12 , 38 ] our results present an evidence against the strong founder effect among Russian FH patients, and it is likely that in the St.-Petersburg area there is as much genetic heterogeneity as in most other areas of the world. Competing interests The author(s) declare that they have no competing interests. Authors' contributions FMZ, MYM, VIG, PHN, GGN, AS performed cloning, SSCP, sequencing and familial analysis, BML, VOK, DD, ADD selected patients, VBV and OF participated in the study design and coordination. All authors read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC551615.xml
497040	Integrated web service for improving alignment quality based on segments comparison	Background Defining blocks forming the global protein structure on the basis of local structural regularity is a very fruitful idea, extensively used in description, and prediction of structure from only sequence information. Over many years the secondary structure elements were used as available building blocks with great success. Specially prepared sets of possible structural motifs can be used to describe similarity between very distant, non-homologous proteins. The reason for utilizing the structural information in the description of proteins is straightforward. Structural comparison is able to detect approximately twice as many distant relationships as sequence comparison at the same error rate. Results Here we provide a new fragment library for Local Structure Segment (LSS) prediction called FRAGlib which is integrated with a previously described segment alignment algorithm SEA. A joined FRAGlib/SEA server provides easy access to both algorithms, allowing a one stop alignment service using a novel approach to protein sequence alignment based on a network matching approach. The FRAGlib used as secondary structure prediction achieves only 73% accuracy in Q3 measure, but when combined with the SEA alignment, it achieves a significant improvement in pairwise sequence alignment quality, as compared to previous SEA implementation and other public alignment algorithms. The FRAGlib algorithm takes ~2 min. to search over FRAGlib database for a typical query protein with 500 residues. The SEA service align two typical proteins within circa ~5 min. All supplementary materials (detailed results of all the benchmarks, the list of test proteins and the whole fragments library) are available for download on-line at . Conclusions The joined FRAGlib/SEA server will be a valuable tool both for molecular biologists working on protein sequence analysis and for bioinformaticians developing computational methods of structure prediction and alignment of proteins.	Background Protein structure is obviously modular, with similar structural segments, such as alpha helices and beta strands found in unrelated proteins. Such segments, identified from structure, are used extensively in description and analysis of protein structures [ 1 , 2 ]. Several groups have demonstrated that only a small library of segments is sufficient to rebuild experimental protein structures with high accuracy [ 3 ]. Predicted local structure segments (PLSS) are also used in structural prediction, starting from the nearest neighbor approach to secondary structure prediction [ 4 - 6 ]. This idea was later extended and lead to even more successful applications of PLSSs in ab initio structure prediction by Baker and colleagues, who developed a library of sequence-structure motifs called I-sites [ 7 ]. Those motifs are later assembled in a complete protein structure by a program ROSETTA [ 8 ]. Predicted local structure segments are also used in a novel protein alignment algorithm, based on the comparison of PLSSs for two proteins treated as networks and finding a common path through networks describing the two proteins [ 9 ]. The underlying idea in all those approaches is that because global folding constraints can override local preferences, the prediction of structure segments from local sequence is by necessity uncertain. Therefore, instead of trying to predict a correct local structure, all possible local solutions are identified and other constraints (folded structure in Rosetta, or compatible alignment in SEA) are used to identify a globally consistent solution. Prediction of local structure segments can be approached in two different ways. A first possibility, used in most nearest neighbor secondary structure algorithms, is to use a representative set of proteins with known structure as source of structure segments, but without any restrictions on a number or type of segments. In this approach, we don't make any assumptions about the compositions and distributions of segments in the library and this approach can be compared to unsupervised learning approach. In a second approach, used for instance in the I-site method, only segments from a specifically constructed fragment library are used in prediction, thus this approach is similar to supervised learning. Interestingly, some limited tests suggest that the former approach leads to lower prediction accuracy [ 10 ]. The same tests suggested the possibility that different segment libraries could lead to different prediction, and likely, some segment libraries would be better suited to some tasks. Following this observation, we have developed the FRAGlib – a fragment library specifically designed to complement a segment alignment SEA. SEA alignment algorithm was developed previously in our group [ 9 ] and originally used in conjunction with the I-site library. I-site library [ 7 ] was originally developed to be used in ab initio folding predictions and anecdotal evidence suggested that it may not be ideally suited for alignment purposes. In this note we describe a combined FRAGlib/SEA server and first benchmarking results of this method. Implementation Database of Short Fragments FRAGlib is based on the idea of developing a uniform coverage of all known types of local structural regularity with the distribution based on that observed in natural proteins. The collection of segments is constructed using representative set of proteins from the ASTRAL database [ 11 , 12 ]. For each protein in this set, each continuous segment with regular secondary structure, including the flanking residues on both sides, is added to the FRAGlib (see below for details). We do not utilize any further clustering algorithm so our database contains no-unique entries and it is redundant both in terms of structure and sequence information. Local structure is described by the SLSR (Symbolized Local Stuctures Representation) codes consisting of 11 symbols { HGEeBdbLlxc }, each representing a certain backbone dihedral (phi and psi) region [ 7 , 13 ]. Protein local structure is described as a string of local-structure symbols and a local structure segment is defined as a 5–17 amino acid fragment with constant local structural codes. Segments are then extended by two additional residues offset at the beginning, and at the end of a segment. We store all such segments with their sequence, SLSR style local structures representation codes and the homology profile [ 14 , 15 ], derived from that of their parent protein. The library is highly redundant, i.e. there are many segments with the same structural description, but each of the redundant fragments is coming from a different parent protein (or a different part of the same parent protein), therefore it has a different sequence and a different profile associated with it. FRAGlib prediction In a next step, FRAGlib segment library is used to assign local structure segments for a new protein (query) based only on sequence information using a variant of the FFAS profile-profile alignment algorithm [ 16 ]. A profile for the query protein is calculated following the FFAS protocol, then for all possible overlapping segments of length from 7 to 19 amino acids, their profiles are compared to those of the segments from the FRAGlib database and the score of each alignment is calculated using a FFAS-like scalar product of composition vectors at each position. Since the segments being compared have the same length, no dynamic programming alignment is necessary and the score calculation can be highly optimized. As the result of this procedure, each position in the query protein can be assigned to all of the possible LSSs in the database, each with a specific score (see Figure 1 ). Only reduced sets of predicted LSSs, rather arbitrarily limited to the first 20 highest scoring segments are kept for further analysis. This cut-off is the only free parameter of the method, and can be set by user using the Web interface of the server. The Q3 quality of the FRAGlib used as a secondary structure prediction algorithm (data not shown), with the prediction based on the single best scoring segment for each position is 73% on a standard secondary structure prediction benchmark. The Q3 gives percentage of residues predicted correctly as helix, strand, and coil or for all three conformational states. SEA Segment Alignment Approach to Protein Comparison The principal motivation to develop the FRAGlib segment prediction was to further improve the alignment quality for comparing distantly related proteins, which is one of the most important problems in practical application of comparative modeling and fold recognition [ 17 ]. To address this problem, we have previously developed a SEA algorithm, which compares the network of predicted local structure segments (PLSSs) for two proteins using the network matching approach. In a previous paper we have demonstrated that the SEA algorithm, using I-site server for PLSSs prediction and a simple sequence-sequence scoring for segment comparison resulted in alignments better than the FFAS profile-profile alignment algorithm and several other alignment tools. A full description of the SEA algorithm is available in the previous manuscript [ 9 ], so only a brief summary is presented here. Every residue in each of the proteins being aligned is described as a vertex in the graph. Two artificial vertices are added to the very beginning of each protein as a source vertex, and also at the end as a sink vertex. For each PLSS is described as an edge between the vertices representing its first and last positions. For some PLSS protocols, some parts of the protein may not be covered by any predicted segments, so virtual edges are added to all neighbor residues to form a complete, continuous network. Each assembly of connected PLSSs corresponds to a path in this network. In a next step, PLSSs networks of two proteins are compared by the SEA algorithm. For each pair of positions i and j , with position i coming form the first protein and position j from the second protein, all possible segments covering each of the positions must be considered in a combinatorial way and compared to get the optimal similarity score. It is not the sequences or secondary structures at two positions that are compared, but all segments that cover these two positions. This is the main feature of SEA that makes it different from standard sequence pair-wise alignments. The computational complexity of SEA is about O ( NMC 1 C 2), where C 1 and C 2 are the average numbers of segments that cover a position in each protein (the segment coverage). Detailed description of the SEA mathematical algorithm together with benchmarks results obtained using the I-site server calculated PLSSs network can be found elsewhere [ 9 ]. The integrated FRAGlib and SEA server is available at [ 18 ]. The FRAGlib database and segment prediction provides the PLSSs network for each aligned protein, and the SEA algorithm aligns the two networks. On Figure 2 we present the flowchart of the integrated web service. Preliminary benchmarks for the FRAGlib/SEA server and presented below. A full paper on the FRAGlib algorithm is in preparation. Results and Discussion We use here as a benchmark the database of 409 family-level similar pairs [ 19 ]. Each protein pair shares at least one similar domain as identified by SCOP [ 20 ]. Segments coming from the proteins of the same SCOP family as the proteins being compared were removed from the FRAGlib calculated PLSSs network. Further analysis of the SEA results also confirmed that the memorization is not a problem here, as all the SEA alignment are build predominantly from segments that are not locally optimal. To evaluate the improvement we use two measures of alignment quality: the classical root mean square deviation (RMSD) and the shift score [ 1 ]. The shift score measures misalignment between a predicted alignment of two proteins and the reference alignment. The shift score measure ranges from -ε(default as -0.2) to 1.0, where 1.0 means an identical alignment. RMSD is dependent on alignment length and the shift score is dependent on the reference alignment, so both measures are less than perfect in comparing alignments. In our case we use as the reference alignment provided by the CE structural method [ 21 ]. We chose the CE, which is available as a single file executable for various operating systems, as an example of purely structural alignment tool. It is a method for fast calculation of pairwise structure alignments, which aligns two proteins chains using characteristics of their local geometry as defined by vectors between Cα positions. Heuristics are used there in defining a set of optimal paths joining termed aligned fragment pairs with gaps as needed. The path with the best RMSD is subject to dynamic programming in order to achieve an optimal alignment. For specific families of proteins additional characteristics are used to weight the alignment. 'Table 1 [see Additional file 5]' compared the quality of the FRAGlib/SEA (identified as SEA F in the Table) alignment with that of the structural alignment prepared with the CE algorithm [ 21 ] and the SEA algorithm used with I-site segment prediction (SEA I ), SEA algorithm used with the actual (not predicted) local structure segments (SEA T ), local single predicted structures (SEA loc ) and few other publicly available alignment tools. All the results other than the FRAGlib/SEA alignments, as well as alignment quality evaluation, were adopted from the original SEA manuscript [ 9 ]. The results presented in 'Table 1 [see Additional file 5]' show that SEA F significantly improves the alignment quality as compared to all other methods, including SEA I (SEA using I-site prediction), bringing it close to (and in the shift based quality measure actually improving on) the SEA algorithm using the actual structure segments. Conclusions The benchmarks show that SEA with FRAGlib (SEA F ) integrated prediction service better incorporate diversities of local structure predictions over known methods. It produces also more accurate alignments in comparison to SEA I (based on the I-site library), or the SEA with single predicted structures (SEA loc ). Comparing those sequence pairwise alignments we can observe that predicted local structure information seems to improve the alignment qualities. Alignments from SEA using FRAGlib method of describing diversities of local structure prediction have the same quality as alignments using true local structures derived from their known 3D structures SEA T . Availability and requirements An integrated SEA/FRAGlib server is available at [ 18 ]. Both components can be used separately, SEA alignment with arbitrary PLSSs and FRAGlib for other purposes than segment alignment, but the integrated server provides the complete alignment method for comparing pairs of protein sequences using a network matching algorithm. The fragments library prediction method (FRAGlib) is also available as the separate http server at [ 22 ]. The software is freely available to academics. Contact Dariusz Plewczynski darman@bioinfo.pl or Adam Godzik adam@burnham.org for information on obtaining the local copy of a software. Authors' contributions DP designed, implemented, and evaluated the FRAGlib program. The benchmark dataset and programme for aligning two short sequence profiles were provided by LJ. The integration of FRAGlib predictions within SEA network alignment software together with benchmark evaluation of the SEA method was done by YY. AG was responsible for the overall project coordination. All authors have read and approved the final manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC497040.xml
439783	Genomic Analysis of Mouse Retinal Development	The vertebrate retina is comprised of seven major cell types that are generated in overlapping but well-defined intervals. To identify genes that might regulate retinal development, gene expression in the developing retina was profiled at multiple time points using serial analysis of gene expression (SAGE). The expression patterns of 1,051 genes that showed developmentally dynamic expression by SAGE were investigated using in situ hybridization. A molecular atlas of gene expression in the developing and mature retina was thereby constructed, along with a taxonomic classification of developmental gene expression patterns. Genes were identified that label both temporal and spatial subsets of mitotic progenitor cells. For each developing and mature major retinal cell type, genes selectively expressed in that cell type were identified. The gene expression profiles of retinal Müller glia and mitotic progenitor cells were found to be highly similar, suggesting that Müller glia might serve to produce multiple retinal cell types under the right conditions. In addition, multiple transcripts that were evolutionarily conserved that did not appear to encode open reading frames of more than 100 amino acids in length (“noncoding RNAs”) were found to be dynamically and specifically expressed in developing and mature retinal cell types. Finally, many photoreceptor-enriched genes that mapped to chromosomal intervals containing retinal disease genes were identified. These data serve as a starting point for functional investigations of the roles of these genes in retinal development and physiology.	Introduction The vertebrate retina is a model system for studying both the development and function of the central nervous system (CNS). Only six major types of neurons develop within the retina, along with a single type of glial cell ( Rodieck 1998 ). These cells are readily distinguished from one another by morphology and laminar position within the retina. Birthdating studies have shown that retinal cell types are generated in overlapping intervals, with ganglion cells, cone photoreceptors, amacrine cells, and horizontal cells generated prior to birth, and bipolar neurons and Müller glia generated after birth in mice ( Sidman 1961 ; Young 1985 a, 1985 b). Rod photoreceptors, the most abundant retinal cell type in the retina, are born both pre- and postnatally, with a peak of genesis coincident with the day of birth in the mouse. These birthdating studies, together with heterochronic coculture experiments ( Belliveau and Cepko 1999 ; Belliveau et al. 2000 ; Rappaport et al. 2001 ), heterochronic transplantation ( Rappaport et al. 2001 ), and lineage analysis ( Turner and Cepko 1987 ; Holt et al. 1988 ; Wetts and Fraser 1988 ; Turner et al. 1990 ), have given rise to the competence model of retinal cell fate specification ( Cepko et al. 1996 ). The competence model states that the intrinsic ability of mitotic retinal progenitor cells to produce a particular cell fate changes continually through development. A cell produces only a single fate, or a subset of fates, at any one time even though lineage analysis has shown that most retinal progenitors have the potential to produce many or all fates over the entire period of retinal development. Interestingly, even at one time in development, retinal progenitor cells show heterogeneity in their developmental competence ( Alexiades and Cepko 1997 ; Belliveau and Cepko 1999 ; Belliveau et al. 2000 ; Rapaport et al. 2001 ). In addition to the contribution of intrinsic determinants of cell fate specification, the fates chosen by the daughters of a retinal progenitor may be influenced by extrinsic factors ( Watanabe and Raff 1990 ; Altschuler et al. 1993 ; Kelley et al. 1994 ; Levine et al. 1997 , 2000 ; Belliveau and Cepko 1999 ; Young and Cepko 2004 ). Finally, certain aspects of retinal cell fate choice, such as the specification of at least some rod and bipolar cells, appear to occur in postmitotic cells ( Ezzeddine et al. 1997 ). Although the competence model was formulated to explain cell fate choice in the retina, it is clear that cell specification in many other regions of the developing nervous system—including neural crest ( Selleck and Bronner-Fraser 1996 ), spinal cord ( Ericson et al. 1996 ), and cerebral cortex ( McConnell 1988 ; Qian et al. 2000 )—involve changes in progenitor competence over time, frequently resulting in altered sensitivity to extrinsic factors. The model of temporal changes in competence is strongly supported by recent elegant studies of Drosophila CNS development ( Isshiki et al. 2001 ; Pearson and Doe 2003 ), where a temporal order of transcription factor expression was found to set the context of cell fate determination. The fundamental similarity among these systems nonetheless accommodates mechanistic differences. The situation in the retina, where early progenitor cells cannot be induced to adopt late fates and vice versa (although see James et al. [2003] for a possible exception to this rule), is distinct from the progressive developmental restriction that is seen in the cerebral cortex, where early cortical progenitor cells are competent to generate cells of upper (late-born) and lower (early-born) layers of the cortex, but become restricted to generating only late-born fates as development proceeds ( Desai and McConnell 2000 ). It is not known what genes mediate changes in progenitor competence during retinal development. Likewise, it is not known to what extent individual retinal progenitor cells from a single time point differ in their developmental competence from one another, although a few genes that are expressed in distinct subsets of progenitor cells have been found ( Austin et al. 1995 ; Matter et al. 1995 ; Alexiades and Cepko 1997 ; Dyer and Cepko 2000 a; Brown et al. 2001 ; Wang et al. 2001 ). Moreover, the genes that regulate the differentiation of any retinal cell type following commitment to a specific fate are generally poorly understood, although a number of transcription factors such as Crx, Nrl, and NR2E3 ( Chen et al. 1997 ; Furukawa et al. 1997 a; Haider et al. 2001 ; Mears et al. 2001 ) are clearly important in rod development. Unbiased, comprehensive expression profiling studies offer the possibility of identifying the molecular components and networks underlying these processes, as well as revealing target genes involved in intermediate and terminal differentiation of individual retinal cell types. We have used serial analysis of gene expression (SAGE) to profile gene expression during the development of the mouse retina ( Blackshaw et al. 2001 ). SAGE, which provides an unbiased and nearly comprehensive readout of gene expression, is conceptually very much like expressed sequence tag (EST) sequencing, with the difference being that concatenated libraries of short sequence tags derived from each cDNA found in the sample of interest are sequenced ( Velculescu et al. 1995 ). By identifying genes that show dynamic expression via SAGE and testing the cellular expression of these genes via in situ hybridization (ISH), we can identify genes that potentially regulate proliferation, cell fate determination, and cell differentiation. Furthermore, by examining SAGE libraries made from adult tissue, genes that are specifically expressed in mature cell types can be identified. By employing both SAGE-based expression profiling and large-scale ISH analysis to determine cellular expression of developmentally dynamic transcripts, we aim to combine the strengths of these two approaches and obtain a detailed picture of molecular events taking place during development of the retina. The laminar structure of the retina, which allows identification of the major cell types expressing a transcript under examination, makes large-scale ISH particularly informative relative to many other regions of the nervous system. Results/Discussion Summary of SAGE Data SAGE was conducted on mouse retinal tissue taken at 2-d intervals from near the start of neurogenesis at embryonic day 12.5 (E12.5) to nearly the end of neurogenesis at postnatal day 6.5 (P6.5). In addition, libraries were made from P10 wild-type mice and the adult retina. Previously generated SAGE data from the microdissected outer nuclear layer (ONL) of the retina, which comprises roughly 97% rod photoreceptors, from retinal tissue from mice that were deficient for Crx (littermates of the wild-type P10 mice), and from adult hypothalamus were also incorporated into the analysis ( Blackshaw et al. 2001 ). All of these libraries were sequenced to a depth of 50,000–60,000 SAGE tags each 14 bp long. Table S1 lists the number of distinct tags found in the 12 retina1 libraries and their abundance levels, along with the number of tags that do not match any known transcript. While 10% of all unique tags found twice or more in the 12 libraries did not correspond to an identified transcript, only 3% of the tags found five times or more did not match a known transcript ( Table S1 ). Table S2 lists all individual tag levels in each of these retinal libraries, along with data from a number of other publicly available nonretinal mouse libraries. We have also created a database, accessible at http://134.174.53.82/Cepko/ , that is searchable by gene name, SAGE tag sequence, accession number, genome location, or UniGene number. It displays all SAGE tags and their levels, as well as ISH images (see below). The accuracy of the SAGE data was assessed by comparing the 15,268 SAGE tags from E14.5 retina to an unnormalized and unsubtracted set of 15,268 ESTs generated by another research group from E14.5 mouse retina of a different strain ( Mu et al. 2001 ). An r-value of 0.65 (see Figure S1 ) was obtained that compares well with SAGE expression profiles obtained in similar tissues but from different individuals that were not strain-matched ( Blackshaw et al. 2003 ). Analysis of SAGE Tag Expression Patterns in Developing Retina Using Cluster Analysis In order to determine whether the temporal pattern of a gene's expression during retinal development might predict its cellular site of expression or its molecular function, clusters of coexpressed genes were assembled. The ten libraries obtained from wild-type total retina were analyzed by cluster analysis using a new Poisson model–based k -means algorithm designed specifically for SAGE data ( Cai et al. 2004 ) (see Materials and Methods for a full description of the algorithm and the protocols used). The results for a 24-cluster analysis are shown graphically in Figure 1 . Table 1 provides a list of previously characterized genes corresponding to tags within these clusters, the number of genes associated with tags within each cluster that were tested via ISH, and select functional categories of genes that were enriched in specific clusters. Table S3 lists all SAGE tags used in the analysis and their corresponding cluster assignments. Figure 1 Median Plot of SAGE Tag K- Means Cluster Analysis Using 24 Clusters Tags present at greater than 0.1% in one or more of the ten wild-type total retina libraries are considered. SAGE libraries are plotted on the x-axis, and tag abundance, plotted as a fraction of the total tags for a gene in the library in question, is shown on the y-axis. A full list of tags and their abundance levels used for the analysis is detailed in Table S3 . Table 1 Summary of SAGE Tag K- Means Cluster Data Tags present at greater than 0.1% in one or more of the ten wild-type total retina libraries were considered. The number of SAGE tags in each cluster is shown, along with the number and percentage of SAGE tags in each cluster that match genes whose expression was examined by ISH in developing retina. Selected genes that were previously examined in the context of retinal development are indicated. P -values for GO categories that are overrepresented in individual clusters were calculated using EASE ( Hosack et al. 2003 ) and represent raw EASE scores for the categories in question Virtually every gene previously reported to regulate retinal development was detected in this analysis and showed dynamic expression during development. Several of these transcripts were found at high levels during their period of peak expression. For instance, NeuroD1 —which regulates rod photoreceptor survival, as well as possibly rod differentiation ( Morrow et al. 1999 ; Wang et al. 2001 )—makes up 0.34% of all retinal mRNA at P4.5. In the case of genes previously shown to be required for production of certain cell types in the developing retina, such as Ath5 and Chx10 —which are required for ganglion cell and bipolar neurons , respectively ( Burmeister et al. 1996 ; Morrow et al. 1999 ; Brown et al. 2001 ; Wang et al. 2001 )—peak expression typically occurred around or just after the peak time of exit from mitosis for that cell type. Certain functional categories of genes were highly overrepresented in a number of SAGE tag clusters. Ribosomal proteins, which typically showed higher expression early in development, were highly enriched in clusters 5, 9, 10, 15, and 23 ( Table 1 )—clusters that also were enriched for cell cycle regulators (particularly clusters 10 and 23). Mitochondrial proteins, by contrast, were concentrated in clusters 4 and 5. Cluster 2 consisted entirely of crystallins, which may be due to contamination by lens tissue in the E12.5 and P0.5 libraries. Phototransduction genes, on the other hand, were found to be concentrated in the late-onset clusters 1, 21, 22, and 24. Genes representing a number of other functional categories also were enriched in specific clusters, although the reasons in these cases are not clear. Examples of this include the concentration of genes involved in RNA processing in clusters 6 and 7, genes coding membrane transporters in cluster 10, and genes that are involved in vesicle-mediated transport in cluster 20. Large-Scale ISH of Dynamically Expressed Genes Genes identified by SAGE were chosen for analysis via ISH by focusing on genes that showed dynamic expression by k- means cluster analysis using Euclidean distance, and some degree of retinal enrichment (i.e., genes were expressed at lower levels in nonretinal SAGE libraries—see Table S2 ). Within this data set, genes whose presumptive function suggested that they might regulate cell fate choice (e.g., transcription factors, growth factors and their receptors, etc.) received highest priority for testing, although many genes of unknown function with developmentally dynamic expression also were tested. See Table S4 for the Gene Ontology Consortium (GO) classification of each probe tested. The analysis was restricted to genes represented by at least 0.1% of total SAGE TAGS in at least one of the retinal libraries, so as to control for sampling variability and to allow for ready detection via ISH. (Exceptions were made for a number of transcription factors and other genes of potentially major functional interest.) This abundance threshold was met by 4,133 tags. Probes corresponding to 1,051 of these tags were tested via ISH. This total included the 346 candidate photoreceptor-enriched genes tested in our previous work ( Blackshaw et al. 2001 ), as well as 37 previously characterized retinal genes that served as positive controls for ISH and to allow clarification of cellular expression patterns. Retinal expression was examined at every time point used for SAGE (see Materials and Methods for details). See Table S5 for a full list of the cellular expression data for each probe in the retina, along with the accession number of the cDNA used to generate each probe used for ISH. See also http://134.174.53.82/cepko/ for images of all of the ISH data. Classification of Cellular Gene Expression Patterns in the Developing Retina The laminar structure of the retina makes it relatively straightforward to assign a tentative identity to cells expressing a given gene. During early stages of retinal development, the outer neuroblastic layer (ONBL) consists almost entirely of mitotic progenitor cells, while newborn neurons (mostly consisting of amacrine and ganglion cells) reside in the inner neuroblastic layer (INBL). The position of mitotic progenitors within the ONBL varies depending upon their progress through the cell cycle, with S phase cells being found on the vitreal side of the ONBL near the border with the INBL and M-phase cells being found on the scleral side of the ONBL abutting the retinal pigment epithelium ( Young 1985 a, 1985 b). Around the time of birth, immature photoreceptors occupy the outer portion of the ONBL. They are comingled with mitotic cells of the G2, M, and G1 phases of the cell cycle, while the S phase mitotic progenitors are in the vitreal side of the ONBL. Finally, by P6, most retinal cells occupy their final positions within the retina. Rod and cone photoreceptors occupy the ONL. Bipolar neuron cell bodies occupy the scleral portion of the inner nuclear layer (INL); the cell bodies of Müller glia occupy a strip in the center of the INL; and amacrine cell bodies are found in the vitreal portion of the INL. The ganglion cell layer (GCL) contains both ganglion cells and a displaced amacrine cells. In the developing retina, expression in the scleral and vitreal portions of both the ONBL and INBL were scored separately, along with whether the gene in question was expressed in all or only a subset of cells in the layer in question. In the case of the adult retina, cell identity in wild-type animals could be scored readily by laminar position of the cells expressing the gene of interest ( Rodiek 1998 ), and thus the identity of expressing cells was scored directly. Extracting order from the diversity of gene expression patterns observed in the developing nervous system can be a daunting task. It is not obvious how best to generate a useful taxonomy of these expression patterns. In tackling this problem, we found it useful to classify cellular expression patterns of genes both by eye and by clustering software. Both methods have specific advantages—user classification more readily identifies rare but distinct patterns, while machine-based clustering allows more flexibility with respect to cluster number and appears to better accommodate classification of intermediate patterns. All classifications were based on the location of the ISH signal within the retinal layers over time during development. Table S6 contains the full list of expression patterns generated by visual inspection, and Table S7 has the full list of cellular expression clusters generated by clustering software. See Materials and Methods for more details on how these data were generated. Comparison of the user-annotated and machine-generated clusters demonstrated fairly strong similarities between the two sets of clusters ( Table S8 ), although genes placed in a single category by user annotation were invariably grouped into larger clusters by clustering software. On the other hand, genes in certain large clusters generated by user annotation—such as panretinal, TRAP2 -like, and Nlk -like (see Table S6 )—were dispersed among many clusters in the machine-generated data sets, with placement within particular clusters varying with replicate program runs. Genes in these categories were expressed at some level in most cells of the developing and mature retina. This variability likely reflects the relative lack of specificity of the expression pattern in these clusters. The finding that most of the highly cell-specific clusters identified by user annotation were readily distinguished by the clustering software supports this hypothesis ( Table S8 ). Using SAGE Data to Predict Cellular Expression Patterns in Developing Retina Temporal changes in gene expression as measured by SAGE turn out to be a useful but inexact method of predicting cellular expression patterns of genes within the retina. While no SAGE cluster was invariably associated with a given cellular expression pattern, genes in certain late-onset SAGE clusters (e.g., clusters 1 and 22) were highly likely to be expressed in developing rods. In the case of early-onset gene expression patterns, which would likely be expressed in retinal progenitor cells, comparison to a microarray-based study could be made. Microarray profiling data of 4N progenitor enriched versus 2N cells has led to the identification of a number of these genes as being enriched in 4N progenitor cells ( Livesey et al. 2004 ). These genes were concentrated in a limited number of SAGE tag clusters (particularly clusters 5, 15, and 23), but were largely absent from clusters that showed a perinatal peak in expression (such as cluster 6), which were enriched for genes expressed in developing rods, bipolars, and amacrine cells (see Table S9 for a full breakdown of 4N-enriched genes by SAGE tag cluster). In general, the temporal expression pattern observed in a given SAGE tag cluster was accurately reflected by the ISH data, although precise prediction of cellular expression patterns based on cluster data were not achieved. Clusters that showed postnatal peaks in expression, such as cluster 6, could contain a great diversity of cellular expression patterns, yet still be enriched for genes that showed strong expression in specific cell types that were differentiating. Table S10 , which details the percentage of tags in a given cluster that represent each specific user-annotated expression pattern, can serve as a starting point for predicting the probability that a gene matching a given SAGE tag will show a given cellular expression pattern in the developing retina. The expression clusters—whether generated by user annotation or clustering software—at best represent a lower limit to the number of distinct expression patterns within the developing retina. Although the number of distinct types of cells in the developing retina is not known, it is undoubtedly high ( MacNeil and Masland 1998 ). Particularly when considering genes expressed in subsets of cells in the ONBL, or subsets of developing amacrine cells, the level of resolution of our ISH-based screen does not allow one to distinguish many of the more complex patterns. Techniques such as multiple-probe fluorescence-based ISH ( Levsky et al. 2002 ) and single-cell microarray analysis ( Tietjen et al. 2003 ) will be required to resolve such questions as whether individual cells coexpress genes that display complex expression patterns. One interesting and potentially useful finding from the SAGE cluster data is that genes known to have highly selective cell-specific expression within a single retinal cell type could show different times of onset of expression. For instance, there is heterogeneity in the time of onset of expression among the genes that mediate rod phototransduction, a feature that has previously been reported in ferret retina ( Johnson et al. 2001 ). Phototransduction genes were found in four different clusters (see Table 1 ), with genes such as RPGRIP showing comparatively early onset of expression, followed by the progressively later onset timesof rod arrestin, rhodopsin, and, finally, Gα1 and GCAP1 (see Table S11 for a full list of tags corresponding to these genes). ISH confirmed the accuracy of the SAGE data for these onset times (see Figure S2 ). This heterogeneity of the time of onset of expression is observed for terminal differentiation markers of every cell type studied in the retina, as well as for markers of subsets of mitotic progenitor cells (see http://134.174.53.82/cepko/ for the full set of ISH data). Such profiles could be explored for the possibility of control by cascades of transcription factors. Gene Expression Patterns Define Subsets of Retinal Progenitor Cells Recent studies in systems as diverse as Drosophila neuroblast specification and the specification of neural-crest-derived cells ( Anderson 1999 ; Isshiki et al. 2001 ; Pearson and Doe 2003 ) have demonstrated the role of temporal changes in gene expression in the specification of neural cells. With respect to the retina, the competence model as originally proposed predicted that mitotic progenitor cells would show both temporal changes in gene expression across broad sets of retinal progenitors, and expression of selected genes in specific subsets of progenitor cells at a given time ( Cepko et al. 1996 ). We have identified a number of genes that show temporally restricted expression in early ONBL. By analyzing the expression of a large number of genes that were highly expressed early in development (particularly in SAGE tag clusters 5, 11, and 15), a number of genes that are expressed in broad but temporally restricted subsets of mitotic progenitor cells were identified ( Figure 2 A). sFrp2 RNA was found to be broadly expressed in the ONBL until E16, after which it rapidly decreased, a pattern that corresponded well with its SAGE tag levels. Expression of Fgf15 and Edr RNA was seen to persist longer, but neither was easily detected after P0, at which time both cyclin D1 mRNA—a recognized marker of mitotic progenitor cells in the retina ( Sicinski et al. 1995 ; Ma et al. 1998 )—and BrdU labeling were still readily detectable in the central retina. Edr RNA showed an unusual patchy distribution in the ONBL at P0—a pattern that was not detected for any other gene tested and has not been previously reported. Lhx2, by contrast, was weakly expressed in subsets of cells in the ONBL until P0, when it was dramatically and transiently upregulated throughout the ONBL. Microarray analysis of 4N versus 2N retinal cells at E16 indicates that both sFrp2 and Lhx2 are enriched in 4N mitotic progenitor cells ( Livesey et al. 2004 ). Figure 2 Genes Expressed in Subsets of Mitotic Progenitors (A) Genes expressed in temporally distinct subsets of progenitors. The first column shows relative SAGE tag levels for each gene under consideration. The UniGene identities and common names of the genes in question are Mm.19155/sFrp2, Mm.3904/Fgf15, Mm.142856/Lhx2, Mm.35829/Edr, and Mm.22288/cyclin D1 . The sections for ISH and BrdU shown here were taken from near the center of the retina at the developmental times shown. Mice were albino Swiss Websters except in the case of the adults, which were pigmented C57B/6. See Table S5 for a full list of probes used. Cellular laminae of both the developing and mature retina are indicated with colored bars. All pictures were taken at 200x. The graph plotting the fraction of mitotic cells in the retina adjacent to the BrdU staining is an estimate based on data from both rat and mouse ( Young 1985 a, 1985 b; Alexiades and Cepko. 1996 ). (B) Spatially heterogeneous ONBL. Genes that were expressed in spatial subsets of cells in the prenatal ONBL are shown. The genes shown are Mm.4541/Sox2, Mm.18789/Sox4, Mm.4605/Tbx2, Mm.29067/Mbtd1, Mm.2229/Eya2, Mm.34701/Pum1, Mm.29924/Arl6ip1, Mm.11738/Ark-1, Mm.40321/Pgrmc2, and Mm.22288/cyclin D1 . Sections were from central retina. Cellular laminae of both the developing and mature retina are indicated with colored bars. All pictures were taken at 200x. See Table S5 for a full list of probes used. To further investigate the expression of these genes in mitotic progenitor cells, ISH was performed on dissociated retinal cells in conjunction with 3 H thymidine labeling at E14, E16, and P0 ( Table 2 ). A substantially lower fraction of double-labeled cells for Fgf15 at P0 relative to earlier time points was observed, while sFrp2 labeling was absent at birth and substantially lower at E16 than at E14. Table 2 Fraction of Progenitor Cells Expressing sFRP2 and FGF15 Decreased as Development Proceeded Retinal explants were labeled with 3 H-thymidine for 1 h, and then disociated and placed on slides. ISH was performed and the fraction of cells expressing sFRP2 and FGF15 is indicated, along with the fraction of cells labeled with 3 H-thymidine, and the fraction of 3 H-thymidine-positive cells that were labeled with probe A limited number of genes have previously been reported as expressed in subsets of mitotic retinal progenitor cells, including genes such as Ath5, and have been shown to be required for retinal ganglion cell development ( Brown et al. 2001 ; Wang et al. 2001 ). We identified a large number of genes that showed selective expression at certain times during development in relatively small subsets of cells in the ONBL ( Figure 2 B). These include a large number of known and putative transcription factors, such as Sox2, Sox4, Tbx2, Eya2 and Mbtd1 (a novel polycomb family member), along with many genes of other functional classes. Particularly intriguing is the early and transient expression of Pum1, a mammalian homolog of the pumilio gene, which has been shown to mediate asymmetric mRNA distribution in Drosophila ( Micklem 1995 ). Many of these genes showed highly dynamic expression during development—rapidly shifting their cellular expression patterns in the course of a few days, as in the case of Pum1 and Sox2, or being expressed for only a few days, as in the case of Eya2 and Pgrmc2 . In some cases, these subsets were scattered throughout the ONBL, such as Eya2 at E14, while for other genes, such as Pum1 and Pgrmc2, expression was in only the scleral portion of the ONBL, suggesting that these genes may show strongest expression near M phase in retinal progenitor cells. From these data, it is difficult to determine whether most of these genes were expressed in cycling progenitor cells or cells that have newly exited from mitosis, as these two populations are intermingled in the ONBL. However, microarray analysis of 4N versus 2N cells of the early retina ( Livesey et al. 2004 ) has indicated that a number of these genes, such as Sox2, are enriched in 4N progenitor cells. See Figure S3 for more examples of genes expressed in subsets of ONBL cells and contrast with Figure S4 , which shows genes with broad but selective expression in the ONBL. The genes that are expressed in subsets of presumptive retinal progenitors include a large number of transcription factors (e.g., Sox2, Lhx2, and Eya2 ) as well as signal transduction components. These intrinsically acting factors represent potential candidates for regulating developmental competence and, by analogy with the Drosophila retina, may act combinatorially to help specify cell fate ( Flores et al. 2000 ). Furthermore, a number of genes that are expressed in temporal subsets of progenitor cells encode secreted differentiation factors such as FGF15 and sFRP2 . Since cell fate choice is determined by the interaction of intrinsic properties and extrinsic factors, these genes are good candidate regulators of cell fate determination. Strikingly, the temporal expression profile of very few progenitor-enriched cell cycle genes tracked precisely with the fraction of mitotic cells in the retina. Even many well-established markers of mitotic progenitor cells, such as cyclinD1 and cdk4 were highly expressed until P2.5 and detectably expressed as late as P6.5—long after the fraction of mitotic cells in the retina had decreased drastically ( Figure 2 A). These data imply that expression of these genes frequently persists after the end of mitosis. In addition, one might have predicted that the levels of cell cycle regulators would be highest at the earliest time point analyzed (E12.5), when the percentage of mitotic cells was highest. However, we found that progenitor-enriched genes such as cyclinD1 and cdk4 often had RNA levels that peaked around P0.5. This observation suggests that the number of mRNA molecules per cell for many of the genes that mediate mitotic activity increases as development proceeds. The functional significance of these findings is unclear, although a number of features of retinal progenitor cells change over the course of development, including the length of the cell cycle ( Young 1985 a; Alexiades and Cepko 1996 ) and the probability of producing progeny that are no longer mitotic ( Livesey and Cepko 2001 ). Genes Expressed in Immature Differentiating Retinal Cell Subtypes One characteristic expression pattern of genes likely to be involved in cell fate specification and/or the early steps of the differentiation process is restriction to newly postmitotic cells and cells actively undergoing differentiation. Many of the genes demonstrated to show such expression in developing retina, such as Crx, Nrl, and NR2E3 ( Furukawa et al. 1997 a, 1997 b; Chen et al. 1997 ; Haider et al. 2001 ; Mears et al. 2001 ) have been shown to play an active role in regulating cell differentiation. We have identified genes that are selectively expressed in immature postmitotic retinal cells of every major class, with the exception of cone photoreceptors, greatly expanding the set of genes known to be selectively expressed in immature retinal precursor cells ( Figure 3 ). KIAA0013, an uncharacterized RhoGAP, was found to be expressed exclusively in immature ganglion cells, and only expressed detectably outside in limited subsets of developing neurons, such as Cajal-Retzius cells of the developing cerebral cortex, and the developing thymus. Cdc42GAP was found to be strongly and transiently expressed in newly postmitotic rods, while the leucine zipper transcription factor Zf-1 was expressed in presumptive bipolar cells. Septin 4 was found to be selectively and persistently expressed in developing horizontal cells, while Mm.23916, a novel dual-specificity protein phosphatase, was found to be expressed selectively in immature amacrine cells. Finally Tweety1, an unconventional chloride channel ( Suzuki and Mizuno 2004 ) was strongly expressed in newly postmitotic Müller glia. Along with genes whose cellular expression could be clearly identified visually, a number of genes with strong but transient expression in undefined subsets of cells of the neonatal retina were observed. Expression of these genes persisted after the end of mitosis in the central retina (see Figure 2 A), so at least some of the cells that express them must be postmitotic. Genes in this category include inhibin βB, brain fatty acid binding protein 7, BMP7, the transcription factor Sal3, and the orphan neurotransmitter transporter NTT7 (see Figure S5 ). Figure 3 Precursor Patterns for Major Retinal Cell Types Genes that are selectively expressed in immature subtypes of retinal cells. From the top, the differentiating cell types that express the genes in question are ganglion cells (Mm.45753/KIAA0013), rod photoreceptors (Mm.103742/Cdc42GAP), bipolar cells (Mm.29496/Zf-1), horizontal cells (Mm.2214 /septin 4), amacrine cells (Mm.23916), and Müller glia (Mm.29729/Tweety1). Sections were from central retina. Cellular laminae of both the developing and mature retina are indicated with colored bars. All pictures were taken at 200x. See Table S5 for a full list of probes used. Genes Expressed in Developing Photoreceptor Cells Rod photoreceptors make up 70% of cells in the retina ( Young et al. 1985 b; Jeon et al. 1998 ). The SAGE-derived expression profile of genes selectively expressed in developing rods is thus more comprehensive than that of other cell types. Based on the ISH data and aided by our SAGE study of mature tissue ( Blackshaw et al. 2001 ), as well as previous reports of mutant mice lacking transcription factors known to be important for rod development, a model of a temporal order of transcription factor expression during rod development was made ( Figure 4 ). Transcription factors known to be involved in cell fate specification sometimes show broad expression in mitotic progenitor cells and persistent expression in mature cell types (e.g., Liu et al. 1994 ; Belecky-Adams et al. 1997 ; Livesey and Cepko 2001 ). We observed a number of genes that were expressed in early ONBL from E16 on, with expression persisting in mature photoreceptors, such as Yboxbp4 . A similar pattern were seen for the mouse ortholog of the Drosophila castor gene, though this gene was observed in a more restricted subset of cells in the ONBL at E16, and for the orphan nuclear receptor ERRβ, although this gene had relatively lower expression prenatally and had pronounced expression in an undefined subset of cells in the immature photoreceptor layer during the first postnatal week. Figure 4 Transcription Factor Cascade in Photoreceptor Development Transcription factors that are selectively expressed in developing rods (and possibly cones as well) are shown. The schematic diagram integrates gene expression data from previously identified photoreceptor-enriched transcription factors and from genes explored in this study. The genes shown are Mm.193526/Yboxbp4, Mm.3499/Rax, Mm. 89623/mCas, Mm.1635/PIAS3, and Mm.235550/ERRβ . See Figure S6 for images of the developmental expression patterns of previously characterized transcription factors. Sections were from central retina. Cellular laminae of both the developing and mature retina are indicated with colored bars. All pictures were taken at 200x. See Table S5 for a full list of probes used. In contrast to being expressed in mitotic cells as well as differentiating photoreceptor cells, a number of transcription factors were selectively expressed in postmitotic but immature photoreceptors. The Rax homeodomain factor showed, as has been previously reported ( Furukawa et al. 1997 a; Mathers et al. 1997 ), strong expression in mitotic progenitor cells in the ONBL that vanished with the end of mitosis. However, expression transiently reappeared in immature photoreceptors at P8. This situation is analogous to that seen in a number of other vertebrates, in which a duplication of the ancestral Rax gene has resulted in Rax genes with distinct expression in photoreceptor and progenitor cells ( Chuang et al. 1999 ; Chen and Cepko 2002 ). PIAS3, which encodes a SUMO lyase that directly regulates the activity of a broad subset of transcription factors ( Kotaja et al. 2002 ; Haider et al. 2001 ), was strongly and selectively expressed only in developing photoreceptors, with expression beginning at E18, peaking at P8, and largely fading away in the adult, a pattern that in many respects is reminiscent of Crx (see Figure S6 ). In contrast to these patterns, Nrl and NR2E3 showed no detectable expression prenatally, and showed peak expression around P6. Somewhat surprisingly, the RNAs for many of these transcription factors is enriched in the inner segments of photoreceptors, as are a large fraction of the other photoreceptor-enriched genes characterized in this study, a finding that is in line with our earlier work ( Blackshaw, et al. 2001 ). The functional significance of this remains unclear. In addition to transcription factors, other functional classes of genes, including genes of unknown function, were expressed in developing photoreceptors, with strongest expression typically found in the first postnatal week ( Figure S7 ). In some cases, these genes fall into pathways known to regulate rod differentiation. Both PIAS3 and the multifunctional protein Hrs ( Chung et al. 1997 ; Scoles et al. 2002 ) selectively inhibit STAT3, and thus possibly inhibit the action of ciliary-derived neurotrophic factor, a factor that has been shown to inhibit rod differentiation in rodents ( Ezzeddine et al. 1997 ; Kirsch et al. 1998 ; Schulz-Key et al. 2002 ). Cdc42GAP expression (see Figure 2 ) may mediate the polarization and initiation of outer segment formation taking place in photoreceptors at this time ( Nobes and Hall 1999 ). In other cases, genes newly identified as selectively expressed in developing photoreceptors imply the existence of novel facets of photoreceptor development. The expression of synaptic vesicle protein Cpx2 suggests that developing photoreceptors may be actively secreting some developmentally relevant signal, while the expression of Hrs also potentially suggests high levels of regulated endocytosis and destruction of unknown extracellular proteins ( Lu et al. 2003 ). The expression of the previously uncharacterized tumor necrosis factor family member Tnfsf13 and A20-like signal transduction components such as TRABID and Fln29 suggest an unexplored role for this pathway in normal photoreceptor development. Genes Expressed in Developing Interneurons of the INL Many genes were selectively expressed in the other, nonphotoreceptor retinal cell types during development. A temporal sequence of transcription factors was observed in bipolar cells as they differentiated ( Figure S8 ). The homeodomain factor Lhx4, and the uncharacterized leucine-zipper protein Zf-1 (see Figure 2 ), showed expression at E16 in the ONBL, with expression continuing postnatally and persisting in adult bipolar cells. Zfh4 was expressed in developing amacrine cells and in subsets of cells in the ONBL prior to P4, and was robustly and transiently expressed in bipolar cells, with peak expression at P6. The relatively late-onset Dbp was first seen in the second postnatal week across the INL. Chx10, as has been previously reported ( Liu et al. 1994 ), and Gli5 were broadly expressed across the ONBL prior to P4, at which point they both showed elevated expression in developing bipolar cells. Microarray analysis confirmed that both of these genes are expressed in mitotic progenitor cells ( Livesey et al. 2004 ). Possible downstream targets of these transcription factors include previously uncharacterized cell adhesion molecules such as the Ig-superfamily member Mm.41284, kinases such as Prkcl, and the putative growth factor receptor SEZ-6 . Furthermore, despite the fact that they comprise only 0.3% of the cells in the adult retina, genes that are highly enriched in both developing and mature horizontal cells ( Figure S9 ), such as the GTPase regulator Borg4, were found. Many genes tested by ISH were selectively expressed in developing amacrine cells ( Figure S10 ). The expression patterns were tremendously diverse, a fact that may reflect the reported extensive heterogeneity among amacrine cell subtypes ( MacNeil and Masland 1998 ). Certain genes, such as the kinase Unc51-like-1, ArfGAP, and the orphan G-protein-coupled receptor Mm.6393, were found to be expressed both in immature amacrine cells and in subsets of cells in the ONBL, particularly in the region of the ONBL that comprises the outer or scleral surface, where M phase mitotic progenitor cells are localized. Cytoskeletal-associated kinases such as Unc51-like-1, and small GTPases such as ArfGAP, may play a role in neurite extension or process formation. Additionally, the expression of neuropeptide receptors such as Mm.6393 in the ONBL before mature neural circuits have formed fits with data from other parts of the developing CNS showing early expression of neurotransmitter receptors and suggesting that neurotransmitters may act on mitotic progenitor cells to regulate cell cycle or cell fate specification ( Rueda et al. 2002 ; Ohtani et al. 2003 ). Similarly, recent work from our laboratory on the role of glycine receptors in the formation of rod photoreceptors ( Young and Cepko 2004 ) confirms such predictions for at least one such receptor. Other genes, such as syntrophin-associated kinase and the novel dual-specificity phosphatase Mm.23916, were confined to immature amacrines only. Syntrophin-associated kinase, in particular, may regulate maturation of synaptic connections ( Lumeng et al. 1999 ). Others genes, such as necdin, the basic helix-loop-helix transcription factor Nhlh2, and the novel PLC isoform Mm.215653, showed complex and often biphasic patterns. The Slit receptor robo3 was strongly and transiently expressed in the first postnatal week in a single sublamina within the INBL, perhaps corresponding to a single subtype of developing amacrine cells. A role for Slit-Robo signaling in regulating cortical dendrite maturation has been demonstrated ( Whitford et al. 2002 ), and these data suggest such a mechanism may be at work in regulating subtype-specific amacrine cell laminae formation in the retina. Neuropeptide Y was strongly and transiently expressed in a subset of amacrine and horizontal cells towards the end of the first postnatal week, with expression dropping dramatically in the adult—suggesting a possible role for this factor in the formation of mature retinal circuitry. Finally Mm.41638, which is weakly homologous to a lysosomal membrane protein, was expressed solely in postnatal amacrine cells, though expression remained in a more restricted subset of amacrine cells in the adult. Müller Glia Are Highly Similar to Retinal Progenitor Cells Genes selectively expressed in Müller glia share a number of defining features. Mitotic retinal progenitor cells and Müller glia showed a great degree of transcriptional overlap—far more so than other retinal cells that differentiate postnatally. Of the genes identified as being specifically expressed in Müller glia after the first postnatal week, 68% were found to be enriched in mitotic progenitor cells based on their ISH pattern, in contrast to only 14% of photoreceptor-specific genes ( Figure 5 A). Of the genes identified as enriched in 4N progenitor cells by micorarray analysis ( Livesey et al. 2004 ) that were tested by ISH in adult retina, 43% were enriched in Müller glia, compared to 11% that were enriched in photoreceptors. Figure 5 Müller-Glia-Enriched Genes (A) Müller-glia-enriched genes show stronger expression in retinal progenitors than do genes enriched in other postnatally born cell types. See Materials and Methods for details of how progenitor-enriched and cell-specific expression patterns were determined, and p -values for progenitor-enrichment of genes that are cell type–specific in the mature retina were calculated. Data on 4N-enriched transcripts were obtained from Livesey et al. (2004) . Numbers for each value are as follows. For N, the number of cell-enriched genes, N MG = 86, N Pr = 112, N BC = 21, and N AC = 57. For I, the number of genes that show retinal progenitor-enriched patterns by ISH, I total = 180, I MG = 66, I PR = 15, I BC = 4, and I AC = 8. For M, the number of genes enriched in 4N retinal progenitor cells that were tested by ISH in adult retina, M total = 28, M MG = 12, M PR = 3, M BC = 3, and M AC = 1. , p < 10 −13 ; *, p < 0.0001. (B) Müller-glia-enriched genes show strong expression in mitotic progenitors. The genes shown are: Mm.26062/ADO24, Mm.55143/Dkk3, Mm.5021/DDR1, Mm.35817, Mm.20465/GPCR37, Mm.200608/clusterin, and Mm.22288/cyclin D1 . Sections were from central retina. Cellular laminae of both the developing and mature retina are indicated with colored bars. All pictures were taken at 200x. See Table S5 for a full list of probes used. (C) Dynamic expression of metabolic genes in developing retina. Metabolic enzymes are often selectively expressed in mitotic progenitors and developing Müller glia. The genes shown are Mm.27953/glycine decarboxylase, Mm.9114/mu-crystallin, and Mm.213213/HK-R . Cellular laminae of both the developing and mature retina are indicated with colored bars. Sections were from central retina. All pictures were taken at 200x. See Table S5 for a full list of probes used. Typical expression patterns for Müller-glia-enriched genes are shown Figure 5 B. Genes in this category, such as the negative regulator of Wnt signaling Dkk3, the collagen receptor DDR1, and the endosomal protein AD024, were observed to be strongly and broadly expressed across the ONBL throughout development, though expression in the adult was restricted to Müller glia. Microarray analysis suggests that a number of these genes, including Dkk3 and DDR1, are enriched in 4N mitotic progenitor cells ( Livesey et al. 2004 ). A smaller set of genes, such as Mm.35817, GPCR37, and Tweety1 (see Figure 2 ) were found to be expressed across the ONBL early in development, but showed dramatically and transiently upregulated expression at the end of the first postnatal week as Müller glia began to differentiate. While over two-thirds of Müller-glia-enriched genes showed enriched expression in retinal progenitors relative to other cell types in the developing retina, virtually all Müller-glia-enriched genes were expressed at detectable levels in retinal progenitors (without necessarily being enriched in progenitors). In fact, only two genes that are Müller-specific in the adult— clusterin and carbonic anhydrase 2 —were expressed in mature Müller glia but not detected in mitotic progenitors. However, previous work suggests that carbonic anhydrase 2 may be expressed in retinal progenitors at levels below our ability to detect ( Vardimon et al. 1986 ), and this may be the case for clusterin as well. Additional Müller-glia-enriched genes are shown in Figure S11 . The extensive overlap in gene expression between Müller glia and mitotic progenitor cells raises the question of how closely these two cell types resemble each other at the functional level. Müller glia morphologically resemble mitotic progenitor cells in having apical and basal processes that span the radial dimension of the retina ( Rodiek 1998 )—a feature that is shared with retinal progenitor cells as well as radial glia of the developing brain, a cell type known to be the cortical progenitor cell ( Doetsch 2003 ). Müller glia are one of the last cell types to exit mitosis ( Young 1985 b; Reh and Levine 1998 ), and they are the only cell type in the mature retina that can reenter mitosis following retinal injury ( Dyer and Cepko 2000b ; Vetter and Moore 2001 ). Finally, data from chicken suggest that, at least in some birds, Müller glia can be induced to divide and give rise to some types of retinal neurons for a short period of time near the end of retinal development ( Fischer and Reh 2001 ). The question arises, then, as to whether Müller glia are fundamentally multipotent progenitor cells that are quiescent regarding cell division and the production of neurons ( Morest and Silver 2003 ; Walcott and Provis 2003 ). If they are progenitor cells, they are progenitor cells that have acquired the specialized properties needed for a support role in the mature retina, e.g., neurotransmitter reuptake and structural roles. The few genes that are specifically expressed in mature Müller glia, such as clusterin, may be emblematic of such roles. Misexpression in mature Müller glia of genes that are candidates for regulating neuronal production in the postnatal retina, followed by injury-induced division, offers a potential approach for future therapies that might lead to photoreceptor or ganglion cell replacement in diseased retinas by cells derived from Müller glia. Prominent Expression of Metabolic Enzymes in Developing Müller Glia A second notable feature of genes expressed nearly specifically in developing Müller glia is the highly dynamic and cell-specific expression of a number of metabolic enzymes ( Figure 5 ). The novel hexokinase-related gene HK-R was selectively expressed in developing Müller glia cells, but not in any other cell in the body examined. Mu-crystallin, which does not encode a crystallin in placental mammals but rather an uncharacterized homolog of the bacterial enzyme ornithine cyclodeaminase ( Segovia et al. 1997 ), showed a similar expression pattern in the retina but also was expressed in other developing sensory organs. Glycine decarboxylase was strongly and selectively expressed in retinal progenitor cells, differentiating Müller glia, and to a lesser extent, developing photoreceptors. The reasons for such high enzymatic activity in development is unclear, although some of these genes may have regulatory functions unconnected to their metabolic roles. For instance, mu-crystallin is also a thyroid hormone binding protein ( Vie et al. 1997 ). Such proteins also may regulate the abundance of small molecules that can act as signals that may be relevant for development. For example, glycine levels may be kept low by glycine decarboxylase so that taurine can bind to and activate the glycine receptor to promote rod differentiation ( Young and Cepko 2004 ). These data point to future directions of research examining the intersection of metabolism and development and suggest the usefulness of supplementing gene expression profiling with metabolomic analysis ( Watkins and German 2002 ). Dynamic Expression of Putative Noncoding RNAs in Developing Retina A number of RNA transcripts that do not appear to encode proteins were strongly expressed in the developing retina ( Figure 6 ). These transcripts are typically spliced and polyadenylated, but do not encode evolutionarily conserved open reading frames (ORFs), or any ORFs encoding proteins longer than 100 amino acids, while often showing high similarity at the nucleotide level between mouse and human ( Numata et al. 2003 ). Table S12 provides a list of these transcripts. Putative noncoding transcripts that showed developmentally dynamic expression include retinal noncoding RNA 1 (RNCR1), which was expressed throughout the ONBL during early development and which was later restricted to Müller glia. It was transcribed in a head-to-head fashion, and largely coexpressed, with Six3. This transcript showed extensive alternative splicing, and while one splice form contained a potential ORF of greater than 100 amino acids, no mouse/human conservation of this putative protein was observed, while high similarity was observed at the nucleotide level in other regions of the transcript. RNCR2 , on the other hand, was expressed in a large subset of cells in both the ONBL and INBL prenatally, with expression restricted to the INL and GCL postnatally. ISH signal for RNCR2 was strongly concentrated in what appeared to be nuclear or perinuclear regions of expressing cells. RNCR3 was expressed in a steadily increasing subset of cells in the ONBL from E14 and gradually resolved to an adult pattern that was photoreceptor-enriched but present in the inner retina at lower levels. Figure 6 Noncoding RNAs in Retinal Development A number of presumptive noncoding RNAs are strongly expressed in dynamic subsets of retinal progenitor and precursor cells. The transcripts shown are Mm.150838/ RNCR1, Mm.44854/ RNCR2, and Mm.194050/RNCR3 . Sections were from central retina. Cellular laminae of both the developing and mature retina are indicated with colored bars. All pictures were taken at 200x. See Table S5 for a full list of probes used. Although additional assays are required to conclusively demonstrate that these RNAs do not encode functional proteins, there is precedent for this conclusion from recent genomic work. Large-scale EST sequencing efforts from mouse have uncovered up to several thousand putative spliced transcripts that do not appear to encode for proteins ( Numata et al. 2003 ). Likewise, oligonucleotide array experiments using probes that tile individual human chromosomes at high density report substantial transcription from many regions not predicted to have protein-coding genes ( Kapranov et al. 2002 ; Cawley et al. 2004 ), and suggest that microarray-based expression profiling that uses probes designed only against predicted protein-coding genes may miss a significant fraction of the transcriptome. The functional role of these transcripts is obscure, although noncoding spliced RNAs such as Xist and H19 in mammals and Rox1 and Rox2 in Drosophila have been implicated in a variety of epigenetic processes ( Mattick 2003 ). The possibility that RNCR1 might somehow regulate expression of Six3 or other progenitor-specific transcripts awaits further investigation. Both Xist and Tsix, noncoding RNAs that play a crucial role in X-inactivation, were expressed in subsets of cells in the ONBL and INBL early in development, but were expressed strongly and selectively in the INL around the end of the first postnatal week ( Figure S12 ). This finding is quite surprising, given that photoreceptors and ganglion cells do not express these transcripts and would thus appear to escape X-inactivation. Since genetic evidence suggests that this is not the case for either cell type ( Reese et al. 1999 ), our findings implicate the existence of possibilities such as alternate cell-specific pathways of X-inactivation or dramatic cell-specific variations in Xist levels required to mediate X-inactivation. Expression Profiling and Candidate Gene Analysis Although we have identified a plethora of transcription factors, growth factors, and signal transduction components, the data do not clearly implicate a known signaling pathway as selectively involved in the differentiation of a given cell type within the retina. For example, negative regulators of Wnt signaling were identified, but these genes display a diversity of cellular expression patterns that cloud a simple model for their action. Dkk3 and Nkd1 are expressed broadly in progenitor cells and Müller glia, together with beta-catenin, while sFRP-2 is expressed exclusively in early progenitor cells, and Nlk is expressed strongly in postmitotic but immature cells of the postnatal retina. Another approach to the creation of models of pathways that control retinal development is to combine the ISH analysis of genes identified via SAGE with a candidate gene approach, even for genes not identified by SAGE. For example, we examined the expression of all known regulators of Wnt signaling, all fibroblast growth factor receptors, and all Slit and Robo genes whether or not SAGE tags corresponding to these genes were identified. See Table S5 and http://134.174.53.82/cepko/ for a full list of genes and their expression patterns. Cell-Specific Gene Expression in the Mature Retina Identifies Candidate Retinal Disease Genes A molecular catalog of gene expression in the adult retina was assembled with molecular markers for every major class of retinal cell ( Figure 7 ). The catalog of photoreceptor-enriched genes reported in previous work ( Blackshaw et al. 2001 ) was expanded, and a large number of genes expressed in the inner retina were identified. Some of these include genes that mark subsets of amacrine and ganglion cells. Knowledge of which genes show cell-specific expression in the retina can aid in identifying retinal disease genes. The expression of nearly half of all cloned photoreceptor dystrophy genes is selectively enriched in photoreceptors ( Blackshaw et al. 2001 ), while hereditary optic neuropathies have been suggested to be partially mediated by mutations in ganglion-cell-enriched genes ( Votruba et al. 1998 ). Furthermore, a number of other retinal and anterior segment abnormalities result from mutations in genes that are broadly expressed in retinal progenitor cells ( Hanson et al. 1999 ; Ferda Percin et al. 2000 ). See Table S13 for a full list of the chromosomal locations of the human orthologs of genes examined in this work. This list also contains a full list of mapped but unidentified Mendelian human retinal disease genes and orthologs of photoreceptor-enriched genes identified in this work that lie within those chromosomal intervals. A total of 164 photoreceptor-enriched genes not previously linked to retinal disease were found in chromosomal intervals containing retinal disease loci, representing a total of 42 distinct loci. While photoreceptor-enriched transcripts make up roughly half of all cloned retinal disease genes ( Blackshaw et al. 2001 ), roughly one-third of retinal disease genes are expressed in all cells of the retina, suggesting that it is fruitful to consider such genes when screening candidate disease genes. We find that 22 panretinally expressed genes map within intervals containing unidentified disease genes, representing 16 distinct loci. Figure 7 Catalog of Gene Expression in Adult Retina The most commonly observed patterns of gene expression in the adult retina are indicated. Data are taken from Table S5 and cover all genes examined in the adult retina. Genes are placed in a category corresponding to a single cell type if expression is substantially greater in that cell type than in any of the other cell types examined. Genes are placed in categories corresponding to multiple cell types if expression is approximately equal in more than one cell type. The number of genes expressed in photoreceptors and Müller glia differs somewhat from those used in the analysis shown in Figure 5 A, since the expression of a large number of photoreceptor-enriched genes was not examined prenatally, and a number of Müller-enriched genes were detectable in Müller glia through the end of the second postnatal week, but not in adult retina. AC, amacrine cells; BC, bipolar cells; GC,ganglion cells; HC, horizontal cells; MG, Müller glia; sAC, subset of amacrine cells; sBC, subset of bipolar cells; sGC, subset of ganglion cells Genomic Approaches to Development The retina consists of a number of distinct cell types that are relatively well defined morphologically, as well as molecularly. They undergo differentiation in defined intervals and are found in stereotypical locations within the retina. These characteristics allow a fairly straightforward evaluation of the cell-specific expression of genes within the retina. We have coupled SAGE-based expression profiling with large-scale ISH analysis to obtain an atlas of gene expression for the developing and mature retina. This atlas is useful for many purposes—in particular, providing many candidate genes for studies of retinal development and function. SAGE analysis can be nearly comprehensive ( Velculescu et al. 1995 ), but its sensitivity is limited by the number of tags sequenced, the level of expression of a transcript within a given cell, and the abundance of given cell subtypes within a tissue sample. Thus this analysis detected relatively rare cell-specific transcripts primarily for the abundant rod photoreceptors and their precursors, and for genes broadly expressed in retinal progenitor cells. Nonetheless, the catalog does include some genes selectively expressed even in the rarest cell types, such as the horizontal cells (0.3% of all retinal cells; Jeon et al. 1998 ) and subtypes of ganglion cells, as well as genes expressed selectively in small subsets of cells in the early ONBL. A recent microarray-based study in developing neural crest screened over 90 candidate genes via ISH ( Gammill and Bronner-Fraser 2002 ), and a recent study using serial stages of embryonic Drosophila has analyzed hundreds of genes by such methods ( Tomancak et al. 2002 ). However, while a number of recent studies have used microarray analysis to profile developing neural tissue, large-scale ISH-based validation of genes identified as being expressed in developing CNS by such expression profiling has not yet been conducted. Large-scale ISH studies enhance our ability to interpret expression profiling data, as the precise cellular expression of a gene in heterogeneous tissues of the developing nervous system cannot be inferred reliably from the profiling of bulk tissue. Other considerations underscore the benefits of verifying primary expression data from expression profiling methods by using other approaches. For instance, several studies describing microarray-based expression profiling of similar starting material have obtained contrasting results for sets of differentially regulated genes ( Claridge-Chang et al. 2001 ; McDonald and Rosbash 2001 ; Lin et al. 2002 ; Ivanova et al. 2002 ; Ramalho-Santos et al. 2002 ). These may result from either experimental variation among labs or biological variation in gene expression among the samples and individuals tested ( Pritchard et al. 2001 ; Blackshaw et al. 2003 ), but nonetheless suggest that large-scale verification of expression differences by techniques such as quantitative RT-PCR or ISH would aid interpretation of such differences. Studies that rely on large-scale ISH as an initial screen generate vast amounts of data, but typically have been conducted using sets of identified or random cDNAs without using expression screening to preselect genes that show high or dynamic expression in the tissue of interest ( Gawantka et al. 1998 ; Neidhardt et al. 2000 ; Kudoh et al. 2001 ; Thut et al. 2001 ). Using expression profiling to generate a set of candidate genes for large-scale ISH analysis will increase the probability of testing genes that show enriched or dynamic expression in a tissue of interest. Towards a Functional Genomics of Neural Development The data presented here provide the starting point for medium-throughput functional analysis of the role of many genes in retinal development. The use of in vivo electroporation ( Matsuda and Cepko 2004 ) and plasmid constructs encoding small inhibitory RNAs delivered by electroporation or retroviruses will make possible medium-throughput gain- and loss-of-function studies of gene function in the retina. The identification of a variety of progenitor subtypes and stage-specific precursor markers will enable a deeper interpretation of such studies. Construction of appropriate Cre lines will allow lineage analysis to determine with precision the mature cell types to which subsets of mitotic progenitor cells or posmitotic precursors give rise. Combining the knowledge of cell-specific transcription factors and cell-specific target genes, together with bioinformatic approaches that take advantage of mammalian genome sequence information in a manner like recent efforts in Drosophila ( Stathopoulos et al. 2002 ), may allow the characterization of the combinatorial code of cis - and trans -acting elements that specify mature neuronal identity. We anticipate that similar approaches are likely to be useful in any region of a developing tissue where birthdating studies have been conducted and cell subtypes can be readily identified based on their spatial localization. Materials and Methods Generation of SAGE libraries Isolation of mouse brain and retinal tissue, as well as construction of all SAGE libraries derived from retinal and hypothalamic tissue, was conducted as previously described ( Blackshaw et al. 2001 ). Publicly available mouse libraries used in the analysis include 3T3 fibroblasts (obtained from http://www.sagenet.org ), P8 cerebellar granule precursor cells maintained in culture for 24 h (GCPcntr; obtained from http://www.ncbi.nlm.nih.gov/SAGE ), P8 cerebellar granule precursor cells maintained in culture and treated with Shh for 24 h (GCP+SHH; obtained from http://www.ncbi.nlm.nih.gov/SAGE ), freshly harvested P8 cerebellar granule precursor cells (GC_P8; obtained from http://www.ncbi.nlm.nih.gov/SAGE ). Libraries from E15 and P1 cerebral cortex were obtained from Gunnersen, et al. 2002 . ). All retinal and hypothalamic SAGE data have been submitted to NCBI, and will be available for download at http://www.ncbi.nlm.nih.gov/SAGE . SAGE data analysis The SAGE 3.0.1 program (courtesy of Victor Velculescu and Ken Kinzler, Johns Hopkins University School of Medicine, Baltimore, Maryland, United States) was used to extract SAGE tags and eliminate duplicate ditags. Identity of SAGE tags was obtained from the National Center for Biotechnology Information (NCBI) “reliable” tag map set for UniGene (available at http://www.ncbi.nlm.nih.gov/SAGE ). UniGene Build 131 of Mus musculus ( http://www.ncbi.nlm.nih.gov/UniGene ) was used for the mappings. In cases where ISH results for genes matching a “reliable” tag did not match the temporal expression profile for the tag in question, along with all cases of unknown tags (i.e., tags which had no “reliable” tag to gene assignment) that were present at greater than 0.1% of total tags in any one SAGE library, the genes were tested via NCBI BLASTN searching ( http://www.ncbi.nlm.nih.gov/BLAST/ ) against the nr and dbest databases, with Expect threshold set to 100 ( Karlin and Altschul 1990 ). A tag was considered to match a specific transcript if it corresponded to the 3′-most NlaIII site in a given polyadenylated transcript ( Velculescu et al. 1995 ). If no such match was found, tags matching the 3′-most NlaII sites in 5′ reads of retinal-derived MGC cDNAs ( Strausberg et al. 2002 ) were considered to match those transcripts, in cases where no further 3′ sequence information was available for those ESTs. Each tag representing a gene tested by ISH, moreover, was checked by BLASTN using these parameters to verify the accuracy of the NCBI tag-to-gene matches. Human orthologs of mouse genes were identified through the use of the Homologene data set and verified by BLASTN and/or BLASTX analysis using the NCBI server, or BLAT analysis using the University of California at Santa Cruz genome server ( http://genome.ucsc.edu ). In cases where no curated ortholog was present in the database, BLASTN analysis against nr, dbest, and htgs databases was used to identify transcripts that showed over 85% sequence conservation over 100 bp and did not match any repeat sequence. The University of California at Santa Cruz genome browser using the October 2003 freeze ( http://genome.ucsc.edu/cgi-bin/hgGateway ) was used to determine if any transcripts with no obvious coding sequence mapped within 5 kb of the 3′ end of an identified gene and were transcribed in the sense orientation relative to that gene. If so, these were considered to represent novel 3′ ends of that gene. All other data analysis and curation was conducted with Microsoft Excel and Microsoft Access. Tissue section, ISH, and BrdU staining ISH was conducted as previously described ( Blackshaw et al. 2001 ). For BrdU staining, mice were given a single interperitoneal injection of 37.5 mg/kg BrdU and killed 1 h later. Fresh-frozen sections were used following 15 min fixation in 4% paraformaldehyde. The protocol of BrdU staining was carried out using an anti-BrdU monoclonal antibody (Roche, Basel, Switzerland) and detected using an AP-conjugated secondary antibody, using recommended blocking and washing conditions. Dissociated cell ISH Retinas were dissected from E14.5, E16.5, and P0 mice and cultured for 1 h in DMEM/10% fetal calf serum containing 5 μCi/ml 3 H-thymidine. The labeled retinas were dissociated into single cells by incubating for 30 min at 37 °C in 100 units/ml of papain (Worthington Biochemical, Lakewood, New Jersey, United States) in Hank's balanced salt solution (HBSS) containing 10 mM HEPES (pH 7.6), 2.5 mM cysteine, and 0.5 mM EDTA. The suspensions were then gently triturated and incubated with 0.1 mg/ml DNase I for 10 min at 37 °C. The cells were pelleted, washed twice in HBSS, and plated on polyD-lysine-coated glass slides for 15 min at room temperature. Cells were fixed to the slides in 4% paraformaldehyde for 5 min at room temperature, washed twice in PBS, and dehydrated in 100% methanol. For acetylation, probe incubation, and subsequent washings, the in situ protocol detailed herein for tissue sections was used. A tyramide signal amplification system (TSA Plus, PerkinElmer, Wellesey, Massachusetts, United States) combined with an anti-digoxigenin-HRP antibody (Roche) was used according to the manufacturer's instructions to detect the signal. Autoradiographic processing was performed in emulsion (NTB2, Eastman Kodak, Rochester, New York, United States) exactly as previously described ( Alexiades and Cepko 1996 ). Classification of cellular expression data in retina by user-based classification and cluster analysis Two classification schemes of the patterns of expression over time were developed: human and machine-aided. In the first case, a single observer (S.B.) generated a presumptive minimal classification of expression patterns following visual inspection of each hybridization pattern (see Table S6 for a full list). This subjective classification took into account a relatively informal assessment of signal intensity. This approach yielded a total of 72 distinct patterns, of which 19 contained only a single member. In the second case, laminar expression within the retina was scored on a 0–5 point scale based upon visual inspection for each defined cell type in the prenatal, perinatal, and mature retina, and cluster analysis software was used to perform k- means clustering (using Euclidean distance) of cellular expression patterns (see Table S7 for the full data set). As with the cluster analysis of the SAGE data, in order to determine an optimal minimal number of clusters, the total distance among data points within the clusters of cellular expression data (within cluster dispersion) were plotted for cluster sizes from 10 to 65 over 100 simulations ( Table S14 ) using Euclidean distance measure ( De Hoon et al. 2004 ). Algorithms used for this analysis are available at http://bonsai.ims.u-tokyo.ac.jp/mdehoon/software/cluster/index.html . It was found that at approximately 45 clusters there was a pronounced discontinuity in the rate of change in the distance among points within the cluster, and this was adopted as a tentative minimal number of clusters. Determination of cell-enriched expression in adult retina and retinal progenitor cells For the data presented in Figure 5 A, numerical cellular expression data from Table S7 was used. Transcripts were assayed as enriched in a specific cell type if they showed highest (but not necessarily exclusive) expression in the cell type in question after the first postnatal week of life. Genes enriched in subsets of bipolars or amacrines were treated as bipolar- and amacrine-enriched, respectively. Whether or not a gene showed retinal-progenitor-enriched expression was determined from Table S7 by the following empirical set of criteria, which were found to cover virtually all known retinal-progenitor-enriched genes: early vO/svO or scO/sscO greater than 1, early (scO + sscO + vO + svO) greater than early (scI + sscI + vI + svI), early (vO + svO) greater than or equal to early (scO + sscO), and mid (vO + svO) greater than mid (scO + sscO). (See legend of Table S5 for a key to these abbreviations.) To determine whether genes that are cell type–specific in the adult retina are disproportionately enriched in retinal progenitors (see Figure 5 A), we have used the hypergeometric distribution statistical analysis to compute the probability that a subset of genes of a given size will have a given number of occurrences of the pattern we examine, when chosen randomly from the group of all known genes ( Johnson et al. 1992 ). Cluster analysis of SAGE data Considering the numerous types of transcripts present in a cell or tissue and the small probability of sampling a particular type of transcript at each draw, the number of sampled transcripts of each type is assumed to be approximately Poisson distributed. Statistically, when this actual sampling process is random enough, Poisson would be the most practical and reasonable assumption compared to other probability models. This assumption, with the assumption that each tag is uniquely mapped to a transcript, leads to the probability model used for clustering analysis of SAGE data (below). First, all SAGE tags were assigned at random to k groups. Second, a cluster center, which led to the expected expression pattern of each tag, was calculated for each cluster. Chi-square test statistics were used to measure the distance between the observed expression pattern and the expected expression pattern of a tag in a cluster. Third, using an iterative method, tags were moved between clusters, and intra- and intercluster distances were measured with each move. Tags were allowed to remain in the new cluster only if they were closer to it than to their previous cluster. Fourth, after each move, the expression vectors for each cluster were recalculated. Last, the shuffling proceeded until moving any more tags made the clusters more variable, increasing intracluster distances and decreasing intercluster dissimilarity (see Protocol S1 for full details of the algorithms used, as well as Cai, et al. 2004 for a more detailed discussion of applications of the protocol). To compute optimal values for the number of clusters k, the within-cluster dispersion was computed for increasing values of k . This within-cluster dispersion declined as new clusters were added. We thus looked for the reduction at each step, and observed the rate of change. Discontinuities in the rate of change were taken to indicate that a meaningful cluster number had been obtained, with the lowest number of clusters that showed such a discontinuity being used for analysis ( Hartigan 1975 ; Yeung et al. 2001 ). In order to determine the optimal number of clusters to use in the analysis of the SAGE data, the within-cluster dispersion was determined for a range of ten to 65 clusters over 100 iterations. If certain numbers of clusters gave a better fit to the data, they should show discontinuities in the rate of decrease ( Hartigan 1975 ). It was found that setting the number of k -means clusters at around 25, 40, and 55 showed these features (see Table S15 ) Database construction Data from 21 SAGE libraries and ISH images were gathered and stored in a MySQL relational database ( http://www.mysql.com ). Information on the measurement values for the SAGE libraries and ISH images can be accessed at http://134.174.53.82/cepko/ . The database was developed to provide up-to-date mapping of SAGE tags to UniGene clusters. Since a single sequence tag can represent different genes and, conversely, an individual UniGene cluster can be represented by more than one tag, both “full” and “reliable” tag-to-UniGene mappings ( Lash et al. 2000 ) have been created and can be selected by the user. The cluster assignments and their reliability were obtained from NCBI SAGEmap ( http://www.ncbi.nlm.nih.gov/SAGE ). For the database reported herein, UniGene Build 131 of Mus musculus and Build 164 of Homo sapiens ( http://www.ncbi.nlm.nih.gov/UniGene ) were used for the mappings. However, the database at http://134.174.53.82/cepko/ includes up-to-date mapping data. For each UniGene cluster, all measurement values and ISH images of associated tags are provided. Measurement values can also be segregated and summed up for each library if more than one SAGE tag is mapped to a given UniGene cluster. A plot of measurement values was also created to visualize patterns across the SAGE libraries. Additionally, for each UniGene cluster, links to gene functions using GO, accession numbers for annotated human orthologs, and LocusLink IDs have been provided. Supporting Information Figures S2–S12 show ISH data for genes that show dynamic expression in developing retina. All pictures were obtained from central retina. Cellular laminae of both the developing and mature retina are indicated with colored bars. All pictures were taken at 200x. See Table S5 for a full list of probes used. Figure S1 Comparison of E14.5 EST Versus E14.5 SAGE Data The number of times a gene was observed in a set of 15,268 individual ESTs obtained from E14.5 mouse retina (data obtained from Mu et al. [2001] ) compared to a set of 15,268 individual E14.5 retinal SAGE tags generated in this study. Only genes present at least ten times in the EST data set were considered. (1.7 MB TIF). Click here for additional data file. Figure S2 Heterogeneous Developmental Onset of Phototransduction Gene Expression The genes shown are rod arrestin, PrCdh, Gγ1, rod PDEγ, rhodopsin, peripherin 2, Gα1, and GCAP1 . (26.9 MB TIF). Click here for additional data file. Figure S3 Genes Expressed in Subsets of Cells in Developing ONBL Sections were from central retina. The genes shown are Otx2, RORβ, Yboxbp1, Mm.38347, Mm.11660, BTF3, H2Ax, Ppp1r14b, Grb10, Mm.158631, HMG-AT1, Mm.24141, KIAA1411, Mm.25018, IAP5, and Chaf1b . (25.7 MB TIF). Click here for additional data file. Figure S4 Genes Expressed Broadly in Mitotic Progenitors The genes shown are PDK3, Giα2, β-catenin, LRC8, Nrarp, Foxn4, and HMG-17 . (27.4 MB TIF). Click here for additional data file. Figure S5 Genes Expressed in Undefined Subsets of Progenitors/Precursors The genes shown are FABP7, BMP7, NTT7, Inhibin βB, and Sal3. (20.8 MB TIF). Click here for additional data file. Figure S6 Known Transcription Factors Expressed in Developing Rods These data are shown to allow direct comparison with the data in Figures 4 and S7 . The genes shown are NeuroD1, Crx, Nrl, and NR2E3. (24.9 MB TIF). Click here for additional data file. Figure S7 Genes Expressed in Developing Rods The genes shown are Cpx2, TRABID, Fln29, Mak, Mm.24642, Nlk, Hrs, Tnfsf13, and Arip2. (18.3 MB TIF). Click here for additional data file. Figure S8 Genes Expressed in Developing Bipolar Cells The genes shown are Chx10, Gli5, Dbp, Lhx4, Mm.41284, Prkcl, SEZ-6, and Zfh4. (21.4 MB TIF). Click here for additional data file. Figure S9 Genes Expressed in Developing Horizontal Cells The gene shown is Borg4. (11.0 MB TIF). Click here for additional data file. Figure S10 Genes Expressed in Developing Amacrine Cells The genes shown are Unc-51-like-1, ArfGAP, robo3, necdin, SAK, Mm.6393, Mm.34130, Nhlh2, NPY, Mm.21657, Mm.215653, and Mm.41638 . (19.1 MB TIF). Click here for additional data file. Figure S11 Genes Expressed in Developing Müller Glia The genes shown are KIAA0937, Mm.157502, Slc38a3, Nkd1, Dsp8, carbonic anhydrase 2, and cyclin D1. (40.1 MB TIF). Click here for additional data file. Figure S12 Additional Noncoding RNAs Expressed in Developing Retina The genes shown are MEG3, Xist, and Tsix. (13.6 MB TIF). Click here for additional data file. Protocol S1 Description of Methodology Used for Cluster Analysis of SAGE Tags (52 KB DOC). Click here for additional data file. Table S1 Summary of SAGE Tag Distribution The total cumulative number of tags found at each abundance level in all 12 retinal libraries (i.e., the ten libraries from total retinal of wild-type animals, the library from P10.5 crx −/− animals, and the library from microdissected ONL of adult animals) is shown. The number of tags, and the fraction of total tags, that do not show any reliable match for any gene (data from NCBI) are also shown. (14 KB XLS). Click here for additional data file. Table S2 Full List of Tag Counts in All SAGE Libraries Considered This list includes not only all libraries made from retinal tissue, but also nonretinal SAGE libraries made by this group, and other mouse libraries that are publicly available. Raw, unnormalized tag counts are shown. See Materials and Methods for more details on the SAGE libraries analyzed. (17.9 MB TXT). Click here for additional data file. Table S3 Twenty-Four-Cluster Analysis for SAGE Tags All tag abundance levels were normalized to 100,000. Tags present at greater than 0.1% in one or more of the ten wild-type total retina libraries were considered. The single most probable “reliable” tag-to-gene match ( http://www.ncbi.nlm.nih.gov/SAGE ) is shown, along with the confidence level of that assignment. Mouse UniGene number is shown for each tag-to-gene match, along with LocusLink ID, where available. In each case where a gene was analyzed by ISH in developing retina, that fact is indicated in the final column. In some cases, a gene that matched the tag with a lower confidence level was tested. In these cases, the UniGene number of the gene tested by ISH differs from that of the most probable tag match. (1.0 MB XLS). Click here for additional data file. Table S4 Molecular Function, Biological Process, and Subcellular Compartment GO Data Are Shown for Each Gene Analyzed by ISH in the Retina Gene names and LocusLink IDs for these genes are also shown (225 KB XLS). Click here for additional data file. Table S5 Complete List of Cellular Expression Patterns for Each Probe Tested The SAGE tag matching each gene tested is given, as well as the accession number of the cDNA used to generate each probe used for ISH. Cellular expression is scored on a 0–5 point scale for each time point tested, as well as for E16 embryo and P6 head cut in horizontal section. A, amacrine cells; Ast, astrocytes; B, bipolar cells; Bv, blood vessels; Cb, cerebellum; CM, ciliary margin; CP, cortical plate; Ctx, cerebral cortex; DG, dentate gyrus of hippocampus; DRG, dorsal root ganglia; EGL, external granule layer of developing cerebellum; EOM, extraocular muscles; G, ganglion cells; H, horizontal cells; Hippo, hippocampus; I, inner neuroblastic layer; In, inner nuclear layer; MG, Müller glia; MGE, medial ganglionic eminence; ND, not determined; O, outer neuroblastic layer; OB, olfactory bulb; OE, olfactory epithelium; ORN, olfactory receptor neurons; P, panretinal; PC, Purkinje cells; PNS, peripheral nervous system; Pr, photoreceptors; Pr(is), inner segments of photoreceptors; sA, subset of amacrine cells; sB, subset of bipolar cells; SC, spinal cord; sG, subset of ganglion cells; sI, subset of cells in INBL; sIn, subset of cells in INL; scI, scleral INBL; sscI, subset of cells in scleral INBL; svI, subset of cells in vitreal INBL; sO, subset of cells in outer neuroblastic layer; scO, scleral ONBL; sscO, subsets of cells in scleral ONBL; svO, subset of cells in vitreal ONBL; sPr, subset of photoreceptors; SVZ, subventricular zone; vI, vitreal INBL; vO, vitreal ONBL; VRN, vomeronasal receptor neurons; VZ, ventricular zone. (381 KB XLS). Click here for additional data file. Table S6 User-Curated Cellular Expression Clusters for Genes Tested by ISH in Retina Here, data from Table S5 are summarized such that the predominant cellular expression pattern from early (E12–E18), mid (P0–P4), and late (P6–adult) developing retina is recorded, and genes are grouped into coexpressed clusters by user annotation. The main cell types expressing the gene in the retina over the interval in question are listed, with weaker expression in other cell types being noted in parentheses. Clusters are given a name (after a representative gene) and a unique cluster number, and the presumptive cell types that show greatest expression are listed. Genes for which the full developmental expression profile was not determined are tentatively assigned to clusters that showed the best fit based on two out of three criteria, with tentative assignments being indicated as such (261 KB XLS). Click here for additional data file. Table S7 Numerical Cellular Expression Data Used for Machine-Aided Cluster Analysis of Cellular Expression Patterns of Genes Tested by ISH in Retina To obtain these numbers, data from Table S5 were modified. As in Figure S6 , expression data were summarized for early (E12–E18), mid (P0–P4), and late (P6–adult) developing retina. In cases where cellular expression changed dramatically within one of these three intervals (e.g., expression shifted from INBL to ONBL), these cellular expressions were both entered in the category in question. Genes that were not examined in all three of these time intervals were not considered in this analysis. Cellular expression data, scored on a 0–5 point scale, were then entered for each time point separately in each of the categories used to score retinal cellular expression in Table S5 . (266 KB XLS). Click here for additional data file. Table S8 Comparison of User-Curated Cellular Expression Clusters from Table S6 and a 45-Cluster Machine-Aided Analysis of the Cellular Expression Data from Table S7 The fraction listed notes the fraction of genes in the machine-generated cluster that were found in a given user-curated cellular expression cluster. The presumptive cellular expression pattern of each user-curated cellular expression cluster is also listed (following Table S6 ). (86 KB XLS). Click here for additional data file. Table S9 Comparison of 4N-Enriched Genes from Livesey et al. (2004) and SAGE Cluster Data from Table S3 Shown is the percentage of tags that matched genes enriched in 4N retinal progenitor cells found in a given SAGE tag cluster. (14 KB XLS). Click here for additional data file. Table S10 Comparison of the SAGE Tag Cluster Data from Table S3 and the 72-Cluster Analysis of the User-Curated Cellular Expression Data from Table S6 Values indicate the fraction of all tags found in a given SAGE tag cluster that were found in a specific user-curated cellular expression cluster. The presumptive cellular expression pattern of each cellular expression cluster is also listed (following Table S6 ). (209 KB XLS). Click here for additional data file. Table S11 SAGE Tags Representing the Known Photoreceptor-Specific Genes Analyzed in Figure S2 Tags in each library are expressed as the fraction of all tags that match the gene in question that were found in the ten libraries considered. (15 KB XLS). Click here for additional data file. Table S12 Candidate Noncoding RNAs Analyzed by ISH in This Study The SAGE tag corresponding to the transcript in question is listed, along with UniGene numbers, and accession numbers of the probes used for ISH for each candidate noncoding RNA. P -values for BLASTN and BLASTX mouse/human comparisons are shown. Transcripts that show high BLASTN, but low BLASTX, matches to human may represent the best candidates for noncoding mRNAs of functional importance and are indicated as likely to be genuine noncoding RNAs. NS, not significant. (17 KB XLS). Click here for additional data file. Table S13 Accession Numbers for Full-Length Transcripts for Genes Tested by ISH in This Study, Along with Their Human Orthologs Chromosomal localizations are shown for both the mouse genes and their human orthologs. Genes located within chromosomal intervals containing mapped but uncloned retinal disease genes are indicated by the name of the disease (terminology from Retnet; http://www.sph.uth.tmc.edu/Retnet/disease.htm ). User-curated cellular expression data of the genes in question (derived from Table S6 ) are shown to aid in prioritizing candidate disease genes for further investigation. ND, not determined. (291 KB XLS). Click here for additional data file. Table S14 Average Distance Analysis of Cellular Expression Data from Table S7 The values shown here are the average sum-of-squares within k- means clusters over all variables. Euclidian mean distance–directed clustering is used ( Hartigan 1975 ). The proportional reduction of error (PRE) for each number of clusters is also shown. This measures the ratio of reduction in within-cluster dispersion to the previous within-cluster dispersion ( Hartigan 1975 ). For this analysis, PRE is given by ( Ni − N ( i – 5))/ Ni, where N is the average within-cluster distance and i is cluster number. (14 KB XLS). Click here for additional data file. Table S15 Average Distance Analysis of SAGE Tag Clusters Tags present at greater than 0.1% in one or more of the ten wild-type total retina libraries were considered and were normalized to 100,000 for this analysis. The average sum-of-squares within k- means clusters for each number of clusters is shown. The PRE, given by ( Ni − N ( i – 5))/ Ni, is also shown. (14 KB XLS). Click here for additional data file. Accession Numbers The GenBank ( www.ncbi.nlm.nih.gov ) accession numbers for the genes discussed in this paper are β-catenin (NM_007614), ArfGAP (BC030682), Arip2 (NM_025292), BMP7 (NM_007557), Borg4 (NM_012121), brain fatty acid binding protein 7 (NM_021272), BTF3 (NM_145455), carbonic anhydrase 2 (NM_009801), cdk4 (NM_009870), Chaf1b (NM_028083), Chx10 (NM_007701), Cpx2 (NM_007756), Crx (NM_007770), Dbp (NM_016974), Drosophila castor gene (BC035954), Dsp8 (XM_181424), FABP7 (NM_021272), Fln29 (NM_172275), Foxn4 (NM_148935), GCAP1 (NM_008189), Giα2 (NM_008138), Gli5 (NM_031184), Grb10 (NM_010345), Gα1 (NM_008140), Gγ1 (NM_010314), H2Ax (NM_010436), HMG-17 (NM_016957), HMG-AT1 (NM_016660), Hrs (NM_008244), IAP5 (NM_009689), inhibin βB (BC048845), KIAA0937 (NM_172442), KIAA1411 (NM_026604), Lhx4 (NM_010712), LRC8 (NM_172736), Mak (NM_008547), MEG3 (NM_144513), Mm.103742/Cdc42GAP (NM_020260), Mm.11660 (AK034313) , Mm.11738/Ark-1 (BC005425) , Mm.142856/Lhx2 (NM_010710), Mm.150838/RNCR1 (AK044330), Mm.157502 (NM_026592) , Mm.158631 (XM_132295) , Mm.1635/PIAS3 (NM_018812), Mm.18789/Sox4 (NM_009238), Mm.19155/sFrp2 (NM_009144), Mm.193526/Yboxbp4 (NM_007705), Mm.194050/RNCR3 (AK044422), Mm.200608/clusterin (NM_013492), Mm.20465/GPCR37 (NM_010338), Mm.213213/HK-R (NM_145419), Mm.215653 (NM_183191), Mm.21657 (BC038057), Mm.2214/septin 4 (NM_011129), Mm.22288/cyclin D1 (NM_007631), Mm.2229/Eya2 (NM_010165) , Mm.235550/ERRβ (NM_011934), Mm.23916 (AK009781), Mm.24141 (NM_025615), Mm.24642 (NM_146168), Mm.25018 (BC010304), Mm.26062/AD024 (NM_025565), Mm.27953/glycine decarboxylase (NM_138595), Mm.29067/Mbtd1 (NM_134012), Mm.29496/Zf-1 (AK004085), Mm.29729/Tweety1 (NM_021324), Mm.29924/Arl6ip1 (BC010196) , Mm.34130 (AK012601), Mm.34701/Pum1 (NM_030722), Mm.3499/Rax homeodomain factor (NM_013833), Mm.35817 (NM_145940), Mm.35829/Edr (NM_133362), Mm.38347 (XM_126644), Mm.3904/Fgf15 (NM_008003), Mm.40321/Pgrmc2 (XM_130859), Mm.41284 (NM_153137), Mm.41638 (NM_029530), Mm.44854/RNCR2 (AK028326), Mm.4541/Sox2 (NM_011443), Mm.45753/KIAA0013 (NM_181416), Mm.4605/Tbx2 (NM_009324), Mm.5021/DDR1 (NM_007584), Mm.55143/Dkk3 (NM_015814), Mm.6393 (NM_010045), Mm.89623/mCas ( BC035954) , Mm.9114/mu-crystallin (NM_016669), necdin (NM_010882), NeuroD1 (NM_010894 , Neuropeptide Y (NM_023456), Nhlh2 (NM_178777), Nkd1 (NM_027280), Nlk (NM_008702), NPY (NM_023456), NR2E3 (NM_013708), Nrarp (NM_025980), Nrl (NM_008736), NTT7 (NM_175328), Otx2 (NM_144841), PDK3 (NM_005391), peripherin 2 (NM_008938), Ppp1r14b (NM_008889), PrCdh (NM_130878), Prkcl (NM_008857), RGPRIP (NM_023879), rhodopsin (NM_145383), robo3 (NM_011248), rod arrestin (NM_009118), rod PDEγ (NM_012065), RORβ (NM_146095), SAK (NM_019945), Sal3 (NM_026528), SEZ-6 (NM_021286), Slc38a3 (NM_023805), syntrophin-associated kinase (NM_019945), Tnfsf13 (NM_023517), TRABID (AK005926), Tsix (AF138745), Unc-51-like-1 (NM_009469), Xist (AK011511), Yboxbp1 (NM_011732), and Zfh4 (NM_030708).	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC439783.xml
524507	Classification between normal and tumor tissues based on the pair-wise gene expression ratio	Background Precise classification of cancer types is critically important for early cancer diagnosis and treatment. Numerous efforts have been made to use gene expression profiles to improve precision of tumor classification. However, reliable cancer-related signals are generally lacking. Method Using recent datasets on colon and prostate cancer, a data transformation procedure from single gene expression to pair-wise gene expression ratio is proposed. Making use of the internal consistency of each expression profiling dataset this transformation improves the signal to noise ratio of the dataset and uncovers new relevant cancer-related signals (features). The efficiency in using the transformed dataset to perform normal/tumor classification was investigated using feature partitioning with informative features (gene annotation) as discriminating axes (single gene expression or pair-wise gene expression ratio). Classification results were compared to the original datasets for up to 10-feature model classifiers. Results 82 and 262 genes that have high correlation to tissue phenotype were selected from the colon and prostate datasets respectively. Remarkably, data transformation of the highly noisy expression data successfully led to lower the coefficient of variation (CV) for the within-class samples as well as improved the correlation with tissue phenotypes. The transformed dataset exhibited lower CV when compared to that of single gene expression. In the colon cancer set, the minimum CV decreased from 45.3% to 16.5%. In prostate cancer, comparable CV was achieved with and without transformation. This improvement in CV, coupled with the improved correlation between the pair-wise gene expression ratio and tissue phenotypes, yielded higher classification efficiency, especially with the colon dataset – from 87.1% to 93.5%. Over 90% of the top ten discriminating axes in both datasets showed significant improvement after data transformation. The high classification efficiency achieved suggested that there exist some cancer-related signals in the form of pair-wise gene expression ratio. Conclusion The results from this study indicated that: 1) in the case when the pair-wise expression ratio transformation achieves lower CV and higher correlation to tissue phenotypes, a better classification of tissue type will follow. 2) the comparable classification accuracy achieved after data transformation suggested that pair-wise gene expression ratio between some pairs of genes can identify reliable markers for cancer.	Background Tumor development is a process in which gene expression is modified, causing abnormal cell behaviour [ 1 ]. Many techniques have been developed to identify abnormalities of gene expression, as reflected by abundance of mRNA transcripts between normal and tumor. The completion of the Human Genome Project and advances in DNA-array technology have allowed highly parallel genetic analyses to take place on a genome-wide scale. They have revolutionized the way tumors are studied, and promised to provide a better and more thorough understanding of the underlying mechanisms for tumorigenesis. Eventually, they will lead to more comprehensive diagnosis/prognosis of tumor with more effective therapeutic interventions. Despite its advantages, the DNA-array technology poses three major challenges that render the interpretation of expression data less efficient than expected. Firstly, the gene expression data is inherently variable due to various factors that either depend on biological factors that remain difficult to control (cross-contaminated samples of tumor and normal cells), or depend on difficulties in setting up of the experiment (RNA extraction) [ 2 ]. These drawbacks interfere with the subsequent array analysis aimed to identify reliable markers that best correlate with the tissue phenotypes. Efforts have been devoted to address these drawbacks by incorporating various raw data scaling, data filtering, normalization and improvement of the classifier algorithm [ 3 ]. Promising results have been reported claiming near-perfect classification accuracy [ 4 ]. However, the usually small number of samples per class in most studies and the highly biased cross validation procedures cast doubt on the classification accuracy in terms of their statistical significance [ 5 ]. This statistical constraint creates a further challenge for DNA-array technology where the number of features in arrays is in thousands while tissue samples are available in limited number. This causes high probability for any classification to be correct by chance alone. Thirdly, although it has been recently established that genes segregate into clusters of interacting networks [ 6 ] instead of acting as one single entity, most cancer DNA-array studies have only investigated single gene aberration (up/down-regulated) when comparing tumor expression profiles to their corresponding normal tissue controls. In an interesting study, B∅ and Jonassen tried to circumvent some of these difficulties by investigating genes in pairs. They demonstrated that gene pairs can be used to improve discrimination between different tissue classes [ 7 ]. This idea of studying genes in pairs, or even in higher order clusters, should be explored further to reveal new features of complex expression profiling datasets. In this study, we introduced a novel data transformation meant to investigate relationships between pair-wise gene expression ratios and tissue phenotype within a given experiment. With this procedure, we aimed to discover strong cancer-related signals (features) that exist in the form of pair-wise ratios (or higher order relationship when we extend to N-feature model classifier for N>2) in a given sample, while improving the signal to noise ratio of the dataset by minimizing its coefficient of variation (CV). The underlying concept for adopting pair-wise gene expression ratios as the discriminating axes for tissue type classification is that an experiment is self-consistent (in terms of factors affected either by the biology of the phenomenon of interest, or of the experimental setting, or both). With this approach we could "subtract" correlated variations by considering the sample as a whole, without making inferences such as those needed for normalization. Basically, we avoided studying gene expression in an absolute term because this requires robust normalization method to account for arrays from different experiments, different platforms and different profiling technologies. By resorting to analyze features in the form of ratios, we attempted to minimize the effect of normalization and look for co-varying signals in each experiment. Methods Colon and prostate cancer datasets The 62 colon cancer sample dataset is composed of measurements for 1,988 gene probes, of which 40 were labelled as tumor and 22 were labelled as normal. The samples were collected from patients, their RNAs were extracted and hybridised to Affymetrix Hum6000 arrays. Please refer to paper [ 8 ]. The normalized dataset can be downloaded at . The 102 prostate cancer sample dataset is composed of measurements for 12,600 gene probes, of which 52 were labelled as tumor and 50 were labelled as normal. The samples were collected from patients, their RNAs were extracted and hybridised to Affymetrix U95Av2 arrays. Please refer to paper [ 9 ]. The normalised dataset can be downloaded at . Both datasets were pre-processed to eliminate those probe pairs that showed significant fluctuation in their hybridisation signals (those greater than 3 standard deviation away from the mean for their ESTs, and the probes pairs that showed an overall higher intensity in their mismatch probe cells (MM) than their corresponding perfect match probe cells (PM); these probe pairs indicate non-specific hybridisation by background RNAs). Both datasets used average intensity as quantitative measurements of the level of gene expression. Base-10 logarithmic transformations were performed for each dataset. Initial gene selection For downstream classification analysis, we extracted only the genes whose expression pattern correlated strongly to the tissue phenotype. To achieve this, we first calculated the correlation coefficient r i (Equation 1) for each gene i using the full dataset, and ranked the genes according to their correlation coefficient r i . For the calculation of r , we assigned a number to each tissue phenotype: 1 for normal tissue and 10 for cancer tissue. After obtaining the correlation coefficients for all genes, we used a simple threshold value (\| r \|>0.4) to select the set of cancer-related genes. There were two reasons for set the threshold value at 0.4. When lower thresholds were used, we incorporated many genes that were not known to be cancer-related (data not shown). Furthermore, too many genes will later cause computer tractability problem when we calculate their pair-wise gene expression ratio for each tissue sample and later the N-feature model classifier. At \| r \|>0.4, we were able to account for most of previously known cancer related genes. where V 1 is a vector representing the gene expression pattern for gene #1; V sample is the dichotomous representation of tissues; S V1 and S sample standard deviation of V 1 , V sample ; , are the mean of V 1 , V sample . Transforming the gene expression data to investigate the expression equilibrium between genes pairs The raw expression data within a sample tissue was transformed into measurement of the pair-wise gene expression ratio for any combinatorial pairs of genes. For the 1,988 gene expression intensities for each sample ( e 1 , e 2 ... e 1988 ), there are 1988 C 2 combinations ( e 1 / e 2 , e 1 / e 3 ...) of pair-wise gene expression ratios (Figure 1 ). This transformed matrix is referred to as M . Each row/column corresponds to a specific gene and the entry at the intersection of row X and column Y corresponds to the expression equilibrium between gene X and gene Y. Such matrix has a diagonal entry of value 1 because e 1 /e 1 equals to unity. Feature partitioning method [ 4 ] for classification of normal/tumor tissues using single gene expression Regarding the F eature P artitioning M ethod (FPM), in order to discriminate between the normal/tumor tissues based on specific feature i (single gene expression), the first step is to determine the threshold value, T i , that can optimally splits all the tissue samples into tumor and normal tissue. The FPM algorithm has a recursive version [ 4 ], in which a decision tree depicting the classification rules for tissue samples was generated recursively. Both methods differ in the way T i s are derived. Nonetheless, they are very intuitive and non-parametric in nature. Also, they restrict no priori distribution patterns for features used. We adopted the simple FPM for tissue classification where each feature was treated individually. There are two criteria for deriving a valid threshold value T i for each feature. First, it has to delineate correctly (discriminating efficiency = 100%) the one-dimensional region (R feature_i ) for either all the normal/tumor tissues using all tissue samples. Secondly, it has to minimize the percentage of false prediction for the other tissue type. Take gene #1659 for example. To fulfill the two aforementioned criteria, it was determined that the region greater than 63.7 (R #1659 ) incorporates all the tumor samples (Figure 2 ). It classified correctly all tumors (discriminating efficiency = 100%) with an overall false prediction of 13.9% in the normal set. This was performed repeatedly for all features until all the threshold values ( T i...all features ) were determined. Now, to classify an unknown sample using 2-feature model classifier, a combination of any two features and their corresponding pre-determined threshold values T i s (selected from T i...all features for each dataset) were recruited. The outcome of the tissue class will be determined depending on whether one/both the expression values of the unknown sample fall completely in either the normal/cancer region (R feature_i ). This is to say that if any of the two features from the unknown sample meets the criteria (R feature_i ) to be either normal/tumor tissue type (based on our definition, R feature_i is a region with 100% discriminating efficiency for a specific tissue type), the unknown sample will be assigned to be normal/tumor respectively. This is repeated exhaustively for all possible combinations constituting of any two features. The procedure will be repeated for all tissue samples to evaluate the overall classification accuracy for 2-feature model classifier. In total, we evaluated the classification of tissue samples based on different combinations of N genes and investigated the classifiers up to 10-feature model classifier. Classification of normal/tumor tissues using transformed datasets The classification procedures and the two criteria for determining the threshold value were the same as explained in previous paragraph. The only difference here is that the definition of "feature" refers to pair-wise gene expression ratio derived from lower/upper triangular matrix of M . Take the ratio #1537/#1831 for example. To fulfill the two aforementioned criteria, it was determined that the region greater than 0.755 (R #1537/#1831 ) incorporates all the tumor tissue samples (Figure 2 ). It classifies correctly all tumor tissue samples with a false prediction of 6.4%. This is performed repeatedly for all entries in M until all the threshold values are determined. Now, to classify an unknown sample using 2-feature model classifier, a combination of any two features (pair-wise gene expression ratio) and their corresponding pre-determined threshold values T i s (selected from T i...all features for each dataset) were recruited. The outcome of the tissue class will be determined depending on whether one/both the expression values of the unknown sample fall completely in either the normal or cancer region (R feature_i ). This is to say that if any of the two features (pair-wise gene expression ratio) from the unknown sample meets the criteria (R feature_i ) to be either normal/tumor (based on our definition, R feature_i is a region with 100% discriminating efficiency for a specific tissue type), the unknown sample will be assigned to be normal/tumor respectively. This is repeated exhaustively for all possible combinations constituting of two features. The procedure will be repeated for all tissue samples to evaluate the overall classification accuracy for 2-feature model classifier. In total, we evaluated the classification of tissue samples based on different combinations of N genes and investigated the classifiers up to 10-feature model classifier. Constructing the relationship tree for the top 25 genes We calculated the cross correlation coefficient r (Equation 1) for all pair combinations of the top 25 genes listed in Table 6 and Table 7 . Prior to the construction of a relationship tree for the top 25 genes for colon and prostate cancer, the cross-correlation coefficient was used to construct the pair-wise distance matrix D . Each entry in the pair-wise distance matrix was measured by the value of (1- r ). Each row/column corresponds to a specific gene and an entry at the intersection of row X and column Y corresponds to the distance of gene expression between gene #X and gene #Y. Such matrix has a diagonal entry of value 0. Only the lower/upper triangular matrix of D is required to construct the relationship tree. After obtaining lower/upper triangular matrix of D , the neighbor-joining method (NJ) algorithm was used to construct the relationship tree [ 10 ]. Computer hardware and software A Sun Fire 6800 Server with 24 CPUs (each running with a clock speed of 900 MHz) was employed throughout this study. The computation of correlation coefficient and classification procedures were implemented using the Matlab Technical Programming language (Matlab programs can be downloaded at . Results After initial gene selection, respectively 82 and 262 genes (\| r \| > 0.4) were selected from the colon and prostate dataset for downstream analysis (Table 1 and Table 2 ). Topping the list in both tables were genes that have been found to be either over-expressed/under-expressed in tumors [ 11 ]. The first three genes most correlated to cancer in the colon dataset were heavy chain of non-muscle myosin, human monocyte-derived neutrophil-activating protein (MONAP) and human desmin genes. This agrees with the findings from [ 12 , 13 ] that used other statistical tests ( z -score, t -test) in a comparable analysis. The heavy chain of non-muscle myosin, denoted as the embryonic smooth muscle myosin heavy chain (SMemb), was found to be down-regulated in cancer. It was also determined experimentally to be a target for the protein encoded by the metastasis-related mts-1 gene [ 14 ]. Furthermore, it was demonstrated recently by 5'RACE analysis that heavy chain of non-muscle myosin interacts with ALK genes that have tyrosine kinase activity and oncogenic properties [ 15 ]. The human monocyte-derived neutrophil-activating protein (MONAP, interleukin-8), was second on the list. It was significantly up-regulated in the tumor compared to the normal samples. This protein has been linked to the progression of several human cancer types [ 16 ]. It was believed that over-expression of MONAP plays an important role in tumor angiogenesis and tumor aggression. The human desmin gene is the third on the list, and it was found to be down-regulated in tumor. Interestingly, this gene also showed significantly reduced expression in other cancer types such as the melanoma cell line [ 17 ]. From the prostate dataset, the most cancer-correlated gene is the human hepatoma gene coding for serine protease hepsin. Brief literature search in PubMed showed that hepsin is a well-characterized transmembrane protease that is expressed at high level in tumor. Three separate studies identified hepsin as a significant cancer biomarker that can be used for cancer diagnosis [ 18 ]. The second gene on the list was the human mitochondrial matrix protein P1. This gene has been correlated to different cancer types with consistent up-regulation in tumor [ 13 ]. The third gene is the carcinoma-associated antigen GA733-2, which was among the 216 cancer markers identified by Ernst's group in Germany [ 19 ]. Effect of data transformation on coefficient of variation To date, reliable markers with low coefficient of variation (CV) are generally lacking. Discovering robust cancer marker is crucial for the purpose of successful cancer diagnosis. We investigated the CV between samples after data transformation: the lowest CVs decreased to 16.5% in the colon dataset while it increased to 25.8% for the prostate dataset (Table 3 and Table 4 ). Topping the list for both dataset were the pair-wise gene expression ratio for genes #119/#54 (elongation factor 1-delta and 40S ribosomal protein S24) and #10614/#5871 (zq58b03.r1 Homo sapiens cDNA and nuclear matrix protein NXP2), which revealed informative pair-wise gene interaction in relation with their corresponding tissue phenotypes. They reflected how cell adjusts to their pair-wise product in response to physiological changes. Based on these observations, we found that the relative abundance between the numerator and denominator exhibited a strong mutual dependency, and had strong correlation to tissue phenotype. For pair-wise gene expression ratio #119/#54, the elongation factor 1-delta is involved in a sequence of events during the decoding of mRNA on the ribosome [ 20 ]. For the ratio of #10614/#5871, it corresponds to novel genes that do not yet have known function. A search in the DNA non-redundant (nr) database for gene #10614 yielded 83% DNA identity to a segment on chromosome 9. On the other hand, a search in non-redundant (nr) database for #5871 revealed 72.3% DNA identity to the cDNA of mouse that incorporates proteins involved in chromosome partitioning and cell decision [ 21 ]. Prior to data transformation the lowest coefficients of variations for single gene expression were 45.3% and 24.5% for colon and prostate datasets respectively. When using the data transformation we proposed, significant improvement was achieved in the colon dataset. Interestingly, this was followed by an improved data correlation to the tissue phenotype as well as to the classification efficiency. We did not observe a similar improvement of the CV, data correlation to tissue classes or classification efficiency in the prostate dataset. Correlations of the single gene expression and pair-wise gene expression ratio The distribution of correlation coefficients between genes and tissue phenotypes for the colon and prostate datasets is shown in Figure 3 . The distributions are positively and negatively skewed for both datasets. The two red lines separate genes with \| r \| >0.4 from the bulk (Table 1 and 2 ). They retained respectively 82 and 262 genes from the colon and prostate datasets. To study the possible interaction between pair-wise genes, we estimated the statistical correlation of gene expressions. Both the distributions for the correlation coefficient and the extreme cases are shown in Figures 4 and 5 . Both figures emphasize the true nature of gene-gene co-regulations – a complex biological mechanism, that most often has been over-simplified when we treat the gene expression as an independent variables [ 22 ]. For example, Figure 4 and Figure 5 suggested that the expressions of genes belonging to a common subset are most likely correlated to each other (e.g.: Gene #31 vs #119 in colon cancer ( r = 0.95306) and gene #7775 vs #10749 in prostate cancer ( r = 0.92922)). It should be pointed out that the two humps in the probability density function are not zero-centered, but concentrated at non-zero correlation r . For colon dataset, positive correlation was the dominant type. For prostate dataset, a balanced distribution in their gene correlation was observed. We determined that some improvement in tissue classification is achieved when pair-wise gene expression ratio was used as discriminating axes instead of using a single gene expression (Figure 2 ). The reason is that pair-wise gene expression ratio has higher correlation to tissue phenotype with lower CV (Table 5 ). Gene expression and tissue type correlation Several previous studies have already endeavored to identify correlations between specific gene expression and cancerous transformation [ 4 , 13 , 23 ]. In the present study, we identified several novel target genes that clearly distinguish the two different tissue phenotypes with high discriminating efficiency (>74%) (Table 6 and Table 8 ). Some of those have previously been documented in studies that did not involve expression profiling as cancer related genes (Human monocyte-derived neutrophil-activating protein (MONAP) and Human hepatoma mRNA for serine protease hepsin), others (Human gene for heterogeneous nuclear ribonucleoprotein (hnRNP), P24480 CALGIZZARIN, Human mitochondrial matrix protein P1, Human mRNA for aldose reductase and human adipsin) have not been identified from in-silico studies of tissue DNA-array expression data. The cancer related genes for colon and prostate cancer were ranked according to their discriminating predictive power. The list should provide hints for researchers during selection of molecular target for diagnostic, prognostic or attempts to cure the disease. Overall classification results and accuracies for each N-feature model classifier across two datasets were reported in Table 6 , 7 and 8 . In the following section, we will discuss a few important genes or pair-wise gene expression ratios from Table 6 and Table 7 that resulted in the optimum classification accuracy (Table 8B ). They are the most efficient combination of discriminating axes for classifying tissue types because they delineate correctly all the normal/tumor tissues with the lowest percentage of false prediction. For the sake of brevity, we will discuss three single gene expressions and two pair-wise gene expression ratios from colon cancer. For prostate cancer, two single gene expressions and two pair-wise gene expression ratios will be discussed. For colon cancer single gene expression, three axes for discriminating tissue types are: 1) Human monocyte-derived neutrophil-activating protein (MONAP); 2) Human desmin gene and 3) Human cysteine-rich protein (CRP) gene. Their threshold values were determined to be 62.73, 2787.0 and 749.4 respectively. For colon cancer pair-wise gene expression ratio, the two axes for discriminating tissue types are: 1) #1831/#1537 and 2) #753/#768. Their threshold values were reported to be 1.32 and 1.85 respectively. For prostate cancer individual gene expression, the two axes for discriminating tissue types are: 1) Human hepatoma mRNA for serine protease hepsin and 2) Human adipsin. Their threshold values were reported to be 115.0 and 182.0 respectively. For prostate cancer pair-wise gene expression ratio, the two axes for discriminating tissue types are: 1) #6185/#5840 and 2) #6185/#6749. Their threshold values were reported to be 2.69 and 2.55 respectively. To illustrate graphically the result of tissue classification, two examples, each based on three genes or pair-wise gene expression ratios that altogether yielded the optimum classification efficiency for the prostate cancer are shown (Figure 6 , Figure 7 ). Constructing the relationship tree for top 25 gene for colon and prostate cancer The relationship tree for top 25 genes listed in Table 6 and Table 7 were constructed based on the cross-correlation between gene expressions (Figure 8 ). We employed the established 'neighbor-joining' clustering method [ 10 ] to group different genes based on their correlated expression patterns across all tissue samples (meaning that genes expression that are correlated will appear in the same branch of the clustering tree), using a novel distance measurement to quantify how change in the expression for one gene interfered with that of another gene. The principle of this method is to cluster pairs of operational taxonomic units (OTUs [=neighbors of similar gene expression]) that minimize the total branch length at each stage of clustering of OTUs starting with a star-like tree. Figure 8 revealed two major clusters of genes. The first cluster corresponded to down-regulated genes, the second cluster represented up-regulated genes. Also, the most efficient discriminating axes (feature genes) reside at the basal position for each cluster. In bacteria many genes are co-expressed as single transcription units. This was used as a control study to validate the methodology of grouping genes, we implemented this distance measurement on bacteria gene arrays ( B. subtilis and E. coli ) and successfully determined the co-regulated operon gene structures (supplementary file #1). Discussion Data transformation to investigate pair-wise gene expression ratios As the expression profiling technologies mature, the identification of significant cancer-related signals from noisy datasets (characterized by a high CV) remains a major challenge. In particular, a robust normalization method is critical to ascertain that arrays from two experiments are comparable with minimum noise prior downstream analysis. However, the existing normalization methods pose limitations due to the lack of good models to account for sources of experimental and biological variations [ 24 ]. Hoffmann et al. [ 25 ] employed different normalization methods to analyse the same dataset, and demonstrated that the numbers of genes detected as differentially expressed differed by a huge factor depending on which normalization methods used. The problem is exacerbated further by the presence of different array formats, experimental designs and methods. Here, instead of resolving to single gene expression, that depends heavily on normalization, for tissue classification, we presented a transformation method that uses pair-wise gene expression ratios within the same experiment as the discriminating axes. By doing so, we aimed to minimize the influence of different normalization methods considering that an experiment is self-consistent with the same factors affecting all genes in the same fashion. The rationale is that even when the normalization methods differ between two array experiments, their pair-wise gene expression ratios within the same experiment will remain relatively stable. If reliable cancer-related signal, exist in the form of pair-wise gene expression ratio, were indeed discovered successfully, they will be relatively independent from the normalization method used on a dataset. The improvement in CV (Table 3 ) and overall classification accuracy (Table 7 ) for colon dataset after introduction of data transformation signifies two implications: First, the transformation is able to increase the signal to noise ratio (SNR) of the cancer related signal because the resulted pair-wise gene expression ratios correlate stronger to tissue phenotype. Second, because the pair-wise gene expression ratios are less dispersed than single gene expression, using the pair-wise gene expression ratios to classify tissue types will be much more reliable and accurate (Table 8 ). Despite the benefits mentioned, this data transformation introduced a computational limitation due to the enormous amount of feature combinations to be processed, especially when N-feature model classifiers for N>4 are considered (If 100 features are selected, and 10-feature model classifier is investigated, the search space will be 100 C 10 = 1.731030945644000 × 10 13 different combination of features). As a result, more computation time will be required to search all possibilities. As an example, the discriminating axes that accounted for the optimum accuracy in 1 to 3-feature model classifier are reported in Table 9 . Regarding the high classification accuracy reported in Table 8 , it should be stressed that this was achieved by involving all tissue samples during the derivation of the threshold value, T i , in the feature selection procedure. In other word, instead of adopting the more conservative classification accuracy test where only a subset of tissue samples are used to derive a set of classification criteria (threshold values), we adjusted our methodology to use all tissue samples so that our results are unbiased (when comparing the outcome from single gene and pair-wise gene ratio) and in-line with our objective that is to compare the classification efficiency between single gene and pair-wise gene ratio. Admittedly, we have a noisy dataset whereby selecting a subset of tissue samples that are a representable population for the entire dataset remains a challenge [ 5 ] (given that we have a small and unbalanced dataset, particularly the colon dataset). Eventually, we might run into ambiguous/contradicting results using a different population subset of tissue samples. Furthermore, we might miss important features (single gene expression/ pair-wise gene expression ratio) because of the biased training dataset. By including all tissue samples for both studies (single gene and pair-wise gene ratio), we aimed to derive the most reliable threshold values and classified tissue samples based on them. Since the same methodology was applied for both studies, the comparison of classification efficiency is valid and will reflect how well each feature (single gene and pair-wise gene ratio) can be used to delineate tissue samples. The implication derived from the classification results For colon dataset, three axes for discriminating tissues are: 1) Human monocyte-derived neutrophil-activating protein (MONAP); 2) Human desmin gene and 3) Human cysteine-rich protein (CRP) gene. The association of the first two genes and cancer biology had been discussed earlier. We will discuss the Human cysteine-rich protein gene. The expression and induction of this protein has been associated with protection against DNA damage, oxidative stress and apoptosis [ 26 ]. In the colon dataset, we observed down-regulation of this protein in tumor. This suggested lack of protection against DNA damage. For colon cancer pair-wise gene expression ratio, the two axes for discriminating tissues are: 1) #1831/#1537 and 2) #753/#768. Using these two axes, 98.4% of the tissue samples can be classified correctly. The expression ratio between #1831 (gelsolin precursor) and #1537 (vascular endothelial growth factor) was able to discriminate 93.6% of the total tissue data. The vascular endothelial growth factor was determined recently to be a plausible biomarker for colon cancer [ 27 ]. Gelsolin had been found to suppress tumorigenicity in different cancer samples, including lung, bladder and breast [ 28 ]. When they were used individually as a discriminating axis, they were only able to classify correctly 66.1% and 67.7% of all tissue samples. Furthermore, the expression ratio between #753 (Human cysteine-rich protein) and #768 (the macrophage migration inhibitory factor) was able to discriminate 90.3% of total tissue type. The human cysteine-rich protein was discussed in the previous section. The macrophage migration inhibitory factor (MIF) functions as a pluripotent cytokine involved in broad-spectrum pathophysiological events in association with inflammation and immune responses. Several reports, including ours, have suggested that MIF is also involved in tumorigenesis [ 29 ]. When they were used individually as single discriminating axis, they were only able to classify correctly 83.9% and 66.1% of all tissues. For prostate cancer single gene expression, the two axes for discriminating tissues are: 1) Human hepatoma mRNA for serine protease hepsin, and 2) Human adipsin. The first gene was discussed in the previous paragraph. For the second gene, adipsin had also been suggested by Chow et al. [ 30 ] as a good cancer marker for studying the basic biology of cancer. For prostate cancer pair-wise gene expression ratio, the two axes for discriminating tissues are: 1) #6185/#5840 and 2) #6185/#6749. Using these two axes, all tissue samples can be classified correctly. The expression ratio between #6185 (Human hepatoma mRNA for serine protease hepsin) and #5840 ( Homo sapiens mRNA for KIAA1109 protein) was able to discriminate 92.2% of total tissues. The human hepatoma mRNA for serine protease hepsin had been determined to be an important marker for cancer cell development [ 11 , 18 ]. The KIAA1109 protein is an unknown protein in human chromosome four [ 31 ]. A homology search against the non-redundant databases yielded no significant hit to known genes. When they were used individually as a discriminating axis, they were only able to classify correctly 86.3% and 61.8% of all tissues. On the other hand, the expression ratio between #6185 (Human hepatoma mRNA for serine protease hepsin) and #6749 ( Homo sapiens mRNA for KIAA1055 protein) was able to discriminate 90.10% of total tissues. The human hepatoma mRNA for serine protease hepsin was discussed in the previous section. The KIAA1055 protein is an unknown protein in human chromosome 15 [ 21 , 31 ]. A homology search against the non-redundant databases yielded 40.7% DNA identity to a novel human cDNA that had been found to function as a cancer inhibiting protein [ 21 ]. When they were used individually as a discriminating axis, they were only able to classify correctly 86.3% and 62.8% of all tissues. Conclusion By comparing the tissue classification methods based on the single gene expression and the pair-wise gene expression ratio in two microarray datasets, we reached the following conclusions: 1. The minimum coefficient of variation decreased from 45.33% to 16.53% for colon dataset but increased marginally from 24.54% to 25.78% in prostate dataset. 2. The correlation coefficient, r , of the discriminating axis that correlates maximally to the tissue phenotype improves from 0.63 to 0.79 and 0.71 to 0.75 in colon and prostate dataset respectively. 3. The optimum accuracy for 1-feature model classifier (using single gene or pair-wise gene expression ratio as discriminating axis) improved from 87.1% to 93.55% in colon dataset. In prostate dataset, nine out of the top 10 discriminating axes showed significant improvement. The mean accuracy for 1-gene classifier improved from 76.8% to 91.2% and 75.8% to 81.9% in both datasets. 4. The comparable classification accuracy achieved after data transformation suggested that there exist some cancer-related signals in the form of pair-wise gene expression ratio, especially prominent in the colon dataset. 5. Through the single gene analysis, we identified key biomarkers that agree with the findings by other researchers. In addition, study on gene-to-gene correlation and the classification outcome based on the pair-wise gene expression ratio suggested that genetic network within a cluster of cancer-related genes should be explored further. Competing interests The authors declare that they have no competing interests. Authors' contributions YLY proposed the idea, participated in the design, performed the statistical analysis and wrote the first draft of the manuscript. AD participated in the design and overall coordination of this study as well as in the writing of the manuscript. XWZ participated in the design of the study. YCW, XHW and MTL participated during the revision phase of this study. All authors read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC524507.xml
549072	Remembering George L. Wied, M.D., February 7, 1921-July 25, 2004	A personal tribute to George L. Wied, M.D., a founder of the medical subspecialty, cytopathology, who died July 25, 2004.	Tribute I arrived in Chicago last November for the ASC meeting, and reflexively reached for the phone to let Dr. Wied know that I had arrived, just as I had for each of countless trips to Chicago over the past 25 years. Then I remembered that he was no longer here, having passed away in Salzburg, Austria, on July 25th. But as the meeting unfolded, I realized that George's mark was deeply imprinted on the entire scientific program, and that his legacy was carried by most of the people in attendance. George Papanicolaou is attributed with the discovery of the cytologic method for detecting epithelial tumors. His laboratory at Cornell Medical Center in New York was the culture medium for the discipline, instructing such notable cytologists as Leopold Koss and George Wied. The medical specialty itself, however, owes its success to George Wied, for without him, it is doubtful whether it would have survived the skepticism of most other medical specialists. Dr. Wied had faith in the importance of cytology that was unwavering and his conviction was transmitted to all who followed him around the globe. A survivor of Nazi Germany, he was a fierce defender of individual freedoms, and he translated that zeal into inclusion of all peoples in his vision. He was an instinctual teacher and taught those around him to convey the criteria of those strange cellular samples via the Tutorials of Cytology. His attention to detail was impeccable, a trait that he insisted his faculty emulate. The TOC faculty quickly became internationally recognized experts, and the Tutorials were a successful enterprise for over 40 years. Until cytopathology became a mandatory part of the curriculum of U.S. pathology training programs, the Tutorials were among the few opportunities for physicians to become facile with the discipline. Repeat enrollment was common, as the specialty became more complex and its importance became apparent to the medical profession. For many nations TOC remained the sole source of first-hand and first-rate cytology education. Judging by the multinational backgrounds of presenters at the ASC meeting, and the excellence of the papers, Dr. Wied's goal has most assuredly been achieved. Setting and maintaining a level of excellence was a responsibility that Dr. George Wied eagerly assumed, and expected the rest of us to uphold. It was easy to follow his example, for his charisma infused us with commitment to those eager to learn. In order to validate the professionals involved in the practice, he established the examinations of the International Academy of Cytology. He personally traveled to every venue where the exam was held, usually following either a Tutorial or International Congress of the IAC. He savored each opportunity to bring another disciple into the fold of cytology. But where Dr. Wied had the most fun was in the realm of cytology automation. As I listened Sunday to the presentations of experiences with the "new" imaging systems and their integration into the clinical cytology laboratory, I recalled the numerous conferences devoted to development of computerized scanners. As early as 1951, Dr. Wied recognized that computers would become an essential part of our daily lives, for data management and communication. He also realized that the task of screening slides was work intensive, subjective, and fraught with opportunities for error. If Dr. Wied didn't have the knowledge himself, he immediately reached out to others who did, and drafted them to the cause of automating his specialty. One of those recruits was Peter Bartels, a gifted optical scientist and incredibly creative problem solver. Together they built a team of researchers at the University of Chicago that attacked each obstacle to success like an army of dragon slayers. Rather than seek the glory of discovery alone, Dr. Wied's generous spirit inspired him to organize meetings to discuss problems common to automation, bringing scientists from a variety of disciplines and numerous countries. If a potentially important contributor to the meeting did not have funding to attend, Dr. Wied would offer to support them, often personally. He was most supportive of young scientists and mentored them through their careers. These were usually pleasant meetings, as Dr. Wied's gentle and considerate nature didn't allow personal bickering. But he loved scientific controversy and he encouraged us to challenge each other to the next level of achievement. He fully believed that what was good for one research group was good for the entire profession. He was most distressed when the commercialization of automated scanners led to open fighting among the companies trusted to develop the fruits of so many years of collegial research. He was a frequent member of NIH study sections that reviewed research proposals in automation. He was also known to drop in unexpectedly on a research group thousands of miles from his home, just to see what was happening. He was free with his advice without being condescending. He would never knowingly offend anyone. His constant encouragement and validation of the importance of scientific discovery for this young specialty was manifested through the two journals he founded and edited, ACTA Cytologica in 1957 and Analytical and Quantitative Cytology and Histology in 1979. Recognition for his accomplishments came frequently and from various sources, including an Outstanding Investigator Award from the NIH. But if you were to reach into the heart of George Wied, I wager that he felt that his greatest accomplishment was being father to Kazutaka (George) Wied, son of his wife, Kay. An unlikely parent, Dr. Wied soon became a master at constructing dioramas out of shoe boxes, holding flash cards with German vocabulary words, and generally encouraging young George to treasure his education. Being a musician himself, Dr. Wied patiently tolerated the squawks and squeaks of his young son's efforts to be a violinist. As young George became more accomplished, after dinner performances soon became an expected treat whenever the occasion allowed. Dr. Wied and Kay continued to travel the world even though Dr. Wied's health was becoming increasingly fragile. His final trip could have been written by a romantic novelist. He accepted a speaking engagement in Japan, but instead of a simple round trip to Tokyo, he insisted on an around-the-world ticket, "because it's cheaper!" For one of the stops, he chose to go to Vienna, a city that he loved, to once more savor the delicious schnitzel at the Intercontinental Hotel, site of the Vienna Tutorials. Kay realized that the Salzburg Music Festival was nearby and would be a good place to relax and hear the music that Dr. Wied loved so dearly. Young George, having just graduated from Stanford University, was with them, not always the case. They spent the day at numerous concerts in Salzburg, all dedicated to Czech composers, an uncanny coincidence since Dr. Wied was born in what was then Czechoslovakia. After the concerts, Kay, young George and Dr. Wied spent the evening together, remarking on the beauty of the music. Dr. Wied died in the night, with his beloved family nearby, and across the street from the home of his most admired composer, Wolfgang Amadeus Mozart. There is an old adage, that when a loved one dies, you mourn each time for all the roles they played in your life. If it is a parent or sibling, you mourn once. However, if the person has been multiple influences in your life, you will mourn for each role. For Dr. Wied, many of us will mourn multiple times. For me, he was a teacher, mentor, advisor, friend, and role model. I will mourn many times, but will be comforted by having known him. I have already and will continue to repay his influence by transmitting his ideals to those who follow. His is truly a legacy that deserves to be perpetuated. George Wied was a man of rare breadth and depth, the kind of professional and human for which mankind will eternally thirst. Figure 1 Dr. Wied at the University of Chicago Cytopathology Laboratory, circa 1990.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC549072.xml
516238	We still fail to account for Mendel's observations	Background The present article corrects common textbook accounts of Mendel's experiments by re-establishing what he wrote and how he accounted for his observations. It notes the long-established tests for the validity of any explanations that purport to explain observations obtained by experiment. Application of these tests to Mendel's paper shows that the arguments he used to explain his observations were internally consistent but were, on one crucial issue, implausible. The same tests are applied to the currently accepted explanation for Mendel's observations. Conclusions The currently favoured explanation for Mendel's observations is untenable. It misrepresents Mendel, fails to distinguish between the parameters and the variables of any system of interacting components, its arguments are inconsistent, it repeats the implausibility in Mendel's paper, fails to give a rational explanation for his observed 3:1 trait ratio and cannot explain why this ratio is not always observed in experimental practice. A rational explanation for Mendel's observations is initiated. Readers are challenged to complete the process before a further article appears.	1. Background We all talk, more or less knowingly, about Mendelian genetics. But four questions need to be asked and answered. 1. Do we understand Mendel's work? To judge from nearly all modern accounts of genetics, we do not. Mendel's paper of 1866 has been persistently misrepresented ever since it was rescued from obscurity in 1900. 2. Do we teach our students a rational description of the inheritance of traits? The answer is again no. Why? Because our current depiction of the inheritance of traits or characteristics is based on false statements, inconsistent arguments and an implausible assertion. 3. Does the current description of Mendelian genetics account for his observations of dominant and recessive traits? No, for the reasons given in answering question 2. 4. Do we account rationally for Mendel's observation of a 3(dominant):1(recessive) trait ratio in some but not all of his experiments? The answer is again no. The reasons will become clear in this article and its successor. A survey of the relevant literature for the period from 1900 to 2003 shows that the various misrepresentations of Mendel's first paper [ 1 ] are of long standing. This is not the place to review all the accumulated historical evidence. The present article concentrates on demonstrating that the currently favoured depiction of elementary Mendelian genetics is untenable; it fails to achieve its intended purpose. A change in the concepts and notation for the interpretation (and teaching) of elementary genetics is suggested. There are two long-established tests of the validity of any hypothesis or proposed explanation for the results observed by experiment. The first test asks: Are all the arguments employed consistent, one with all the others? The second test asks: Are all the proposed mechanisms plausible? Could they be confirmed by experiment, i.e. by a "real"experiment or by a logical "thought experiment". Both tests must be passed if the proposed explanations for the observations are to be accepted. If judgement is being passed on work carried out in the distant past, allowance must be made for the availability or lack of availability of tests of plausibility at that time. On the other hand, we should not hesitate to criticise a current explanation that fails tests of plausibility that are now available but were not available in the past. These two tests of validity (consistency and plausibility) will be applied to Mendel's explanation for his observations and to the currently favoured explanation for his observations. We must first re-establish what experiments Mendel performed and what he wrote in his published accounts of these experiments in order to correct the various false textbook descriptions of Mendel's work. For this purpose it is necessary to study authentic reprints of his two papers [ 1 , 2 ]. The first paper is the one we are concerned with here; it was reprinted [ 3 ] and in a version [ 4 ] correcting several type-setting errors that occurred when Mendel's manuscript was set in typescript. The translation into English by Sherwood [ 5 ] avoided several errors in earlier attempts to translate Mendel's Versuche paper [ 1 ]. There may be other sound translations, but Sherwood's version is strongly recommended. It is accurate and also captures Mendel's literary style. 2. Mendel's experiments and his conclusions 2.1. Why did Mendel carry out his experiments? Many earlier biologists had noted the appearance of hybrid plants but their findings did not show how hybrids arose, whether there was any regularity in their occurrence, or how their properties were related to those of their parents. Mendel showed that there was a general rule for the appearance of hybrid plants and that an exact relationship existed between the traits displayed by hybrids and those displayed by their parents. Hence the title of his first paper: Versuche über Pflanzen-hybriden (Investigations on plant hybrids). 2.2. Mendel's preliminary work and his conditions for successful experimentation Mendel recognised five preconditions for success in his experiments on the origin of hybrids: (i) He needed suitable plants for his experiments. He chose Pisum sativum (the edible pea plant) for most of his work because many established varieties were readily available; and because the flowers enclose the reproductive organs, so minimising accidental cross-fertilisation by insect-or air-borne pollen. (ii) Pisum sativum , like all leguminosae, is androgynous. The flowers contain both male (pollen or sperm) and female (germinal or ova) cells and are therefore normally self-fertilising. This provided experimental advantages, as we shall see. (iii) It was necessary to have stocks of true breeding plants for his cross-fertilisation experiments. He therefore spent much time establishing that 22 varieties of edible pea plants were in fact true breeding. He discarded those plants that were not true breeding before starting his experiments on hybridisation. (iv) He had to ensure that any cross-fertilisations were strictly under his control. To achieve this control, he removed all the immature pollen-bearing stamens from a true-breeding pea plant that displayed a particular trait, e.g. green seeds, then transferred pollen to these emasculated flowers from another true breeding pea plant that displayed an alternative form of the same trait (e.g. yellow seeds). (v) Success depended on meticulous enumeration of the occurrence of hybrids, and of alternative traits, in the populations of plants that arose from his cross-and self-fertilisation experiments; and on repetition of each cross-fertilisation and self-fertilisation experiment in order to obtain reliable, average, results. Table 1 reveals the magnitude of Mendel's undertaking and records his observations on the occurrence of hybrids, and of plants displaying either dominant or recessive traits (see further descriptions in the following section). Reciprocal crosses gave the same results; Mendel thus established that male and female sex cells contributed equally to the final outcomes. Table 1 Mendel's novel observations summarised. Mendel demonstrated that crossing parental plants bearing alternative forms ( A ) and ( a ) of any one of seven traits generated a F1 population of plants (not shown) all of which were hybrids ( Aa ). Each of these F1 hybrid plants displayed only one of the two alternative parental traits, defined as the dominating trait ( A ). When these F1 hybrid plants were allowed to self-fertilise, the ratio of dominant to recessive traits in the F2 population was always close to 3:1. Pairs of parental plants Their F2 progeny Dominant traits ( A ) Recessive traits ( a ) Number of F2 plants examined Dominant:recessive trait ratio in the F2 population Green pods Yellow pods 580 2.82:1 Axial flowers Terminal flowers 858 3.14:1 Red flowers White flowers 929 3.15:1 Long stems Short stems 1064 2.84:1 Inflated pods Constricted pods 1181 2.95:1 Round seeds Wrinkled seeds 7324 2.96:1 Yellow seeds Green seeds 8023 3.01:1 2.3. Bateson's notation for successive stages in breeding experiments The following account uses the notation proposed by Bateson [ 6 ] for successive generations arising from sexual reproduction:- P = the original male and female parental generations; F1 = the first filial progeny population arising from crosses between plants of the P generation; F2 = the second filial generation that arises from sexual reproduction by members of the F1 generation – and so on. The advantage of Bateson's notation is that it does not depend on any preconceived ideas about the mechanisms of inheritance of traits during sexual reproduction. It can therefore be used to describe the stages in Mendel's experiments without misrepresenting any of his observations, arguments or conclusions. 2.4. Mendel's initial observations summarised Table 1 shows the results of seven different cross-fertilisations between parental (P) plants displaying alternative forms of the same trait, e.g. red rather than white forms of the trait "flower colour"; all individual plants in the F1 population displayed only one of the two parental trait forms. Also shown are the results observed by Mendel when he allowed these F1 plants to self-fertilise; the ratio of ( A ) form to ( a ) form plants was, in every case, close to 3:1. Mendel also carried out experiments in which he cross-fertilised plants displaying concurrently two or three trait differences, and then recorded the occurrence of each trait in the F1 and F2 generations. These results are not shown here but they were consistent with the findings exemplified in Table 1 . These initial findings led Mendel to a remarkable generalisation and a definition. (i) All plants in the F1 population displayed only one of any two differing trait forms ( A ) and ( a ) displayed by the parental (P) plants. (ii) He defined the trait form that was displayed in the F1 plants as das dominirende Merkmal ( A ) – the dominating trait (A) . He defined the alternative trait form, which did not appear in any of the F1 plants, as das recessivem Merkmal (a) – the recessive trait ( a ). 2.5. Further experiments Mendel now faced the problem of explaining how the 3(dominant):1(recessive) trait ratio arose in the F2 population of plants (Table 1 ). In further experiments on each of the seven crosses shown in Table 1 , he was able to show that those F2 plants he had identified by the symbol ( a ) were 'constant form' (true-breeding) plants; i.e., when they were allowed to self-fertilise, all their F3 progeny displayed the same parental trait ( a ). On the other hand, when F2 plants initially identified by the symbol ( A ) were allowed to self-fertilise some proved to be 'constant form' plants because, when they were allowed to self-fertilise, they produced F3 progeny that again displayed this same parental trait ( A ). But other plants initially identified by the symbol ( A ) in the F2 population were not 'constant form' plants. Some of their F3 progeny did display the original parental trait ( A ). Other plants in the same F3 population displayed the alternative parental trait ( a ). Yet other plants in this F3 population were again not 'constant form' plants. They were like the F1 plants (their "grandparents") and like the F2 parents from which they were immediately derived. When they were allowed to self-fertilise, some of their progeny displayed the ( A ) form, some the ( a ) form of trait and some were again like the F1 plants. The experimental procedures Mendel used to make these distinctions are readily understood by reading a reprint of the original paper or a reliable translation. Given this ability to distinguish, by experiment, between those plants initially designated ( A ) and those now designated ( Aa ), Mendel was able to state the average distribution of trait forms among the plants of the F2 population as (one dominant: two hybrid: one recessive) or, in his notation, (A + 2 Aa + a) ; i.e. the 3:1 trait ratio factored into the proportions 1:2:1. Mendel was now able to add a further generalisation: When F1 plants were allowed to self-fertilise, 1/4 of the F2 population displayed the 'constant form' parental trait (A) that was displayed by the F1 plants, 1/4 displayed the 'constant form' parental trait (a) that did not appear in any of the F1 plants (Table 1 ), and 1/2 were hybrids ( Aa ) that displayed only the dominant trait ( A ) but were not 'constant form' plants. 2.6. Mendel's notation Mendel used upper case and lower case italicised letters throughout his paper to denote, by definition, dominant and recessive traits . Examples have already been given of the use of letters ( A ) and ( a ) when only one trait difference between parental plants was tested (Table 1 ). Mendel made similar use of the letters ( B ) and ( b ), ( C ) and ( c ) when he described experiments in which two or three trait differences were displayed concurrently. For reasons given in Section 2.5 these single letters also designated what Mendel called 'constant forms' of traits. Plants displaying these traits were 'true-breeders'; they were the parental plants he used in cross-fertilisations (Table 1 ). There is one further crucial feature of Mendel's single letter notation for 'constant form' traits. These letters ( A, a, B, b, C, c ) did not represent the structure or composition of the traits. All the traits shown in Table 1 obviously had complex compositions. But, irrespective of such complexity, each dominant trait was denoted by ( A ) and each recessive trait by ( a ) in Table 1 . The traits were what Mendel could see with his own eyes. He distinguished a dominant trait from a recessive trait by qualitative observations. He was not concerned with and did not analyse the structural composition of the traits. The letters ( A, a, B, b, C, c ) represented classes of traits – a dominant class represented by an upper case letter, and a recessive class of trait represented by the corresponding lower case letter (Table 1 ). It is necessary to recognise these facts if a rational explanation for Mendel's observation is to be obtained; and if gross misrepresentations of Mendel's paper are to be detected. Why then did Mendel use a combination of letters (e.g. Aa ) to represent hybrid plants? This will become clear in section 2.7. 2.7. Postulates and arguments; Mendel's explanations of his observations Mendel accounted for the two generalisations (section 2.4) by the following postulates and arguments; they were based on his further experiments (section 2.5):- (1) All the F1 plants were hybrids ( Aa ) in welcher beide Merkmale vereinigt sind – in which both (parental) traits ( A and a ) were united; trait ( a ) was not displayed by these hybrids, so that these hybrids displayed what he had defined as the dominating trait ( A ) only. (2) The traits ( A ) and ( a ) in the F1 hybrids ( Aa ) segregated into traits ( A ) and ( a ) during formation of the male pollen (sperm) cells and also during formation of the female germinal cells (ova). Thus, each pollen cell and each germinal cell carried only one trait – either ( A ) or ( a ) but not both. (3) Fertilisation of one germinal cell by one pollen cell was a random event. (4) When a pollen cell bearing trait ( A ) fertilised a germinal cell bearing the same trait ( A ), all their progeny displayed the trait ( A ). Likewise, when a pollen cell bearing a trait ( a ) fertilised a germinal cell bearing the same trait ( a ), all their progeny displayed the trait ( a ). But when a pollen cell bearing trait ( A ) fertilised a germinal cell bearing the alternative trait ( a ), the resulting plant was the hybrid ( Aa ); if the pollen cell displaying a trait ( a ) fertilised a germinal cell displaying the alternative trait ( A ), the outcome was again a hybrid ( Aa ). In either event, the hybrid ( Aa ) displayed only the dominant trait ( A ). (5) Mendel illustrated these postulates and explanations in a diagram (Figure 1 ) showing the consequences of self-fertilisation of F1 hybrids ( Aa ), given that traits ( A ) and ( a ) in the hybrid ( Aa ) first segregated into individual pollen cells (sperm) and individual germinal cells (ova) before recombining, in random fashion, during formation of the F2 population. The arrows in Figure 1 represent the fertilising event. Figure 1 Mendel's diagrammatic explanation for the formation of the F2 population of plants produced by self-fertilisation of his F1 hybrids. Mendel proposed that F1 hybrids ( Aa ) contained a dominant trait ( A ) that was displayed and a recessive trait ( a ) that was not displayed. Self-fertilisation of F1 hybrids ( Aa ) then involved segregation of the component traits ( A ) and ( a ) into individual male pollen and female germinal cells, as shown in his diagram. Mendel proposed that if a male pollen cell carrying a trait ( A ) fertilised a female germinal cell carrying the same trait ( A ), the progeny would display trait ( A ). He used the analogous argument for the generation of progeny bearing trait ( a ). Only if male and female sex cells carried differing forms of a given trait ( A or a but not both) would the progeny be hybrids ( Aa ). Thus random recombination of the segregated traits during self-fertilisation of hybrids would yield (on average) the F2 population of plants represented by the trait series ( A + 2 Aa + a ) shown below Mendel's original diagram. Note two crucial points:- (i) Mendel observed and recorded the occurrence of traits ( die Merkmale ) or the characters (die Charaktere) in his plants and their seeds, not the mechanisms underpinning these occurrences. These mechanisms could not have been investigated in 1866. (ii) All Mendel's explanations were based solely on observations of the changes in the occurrence of alternative traits in successive populations that arose from cross-or self-fertilisations and back-crosses. 2.8. Comment Mendel was a well-trained scientist [ 7 ], an astute thinker, a careful and systematic experimentalist, an expert hybridiser and an exemplary writer but he was not the first geneticist. That title should go, possibly, to Bateson [ 6 , 8 , 9 ] for advocating Mendel's experimental methods, for showing that Mendel's findings could be repeated in animals, and for emphasising that combination , segregation and recombination of traits during gametogenesis was the most important feature of Mendel's work. Moreover, Bateson realised [[ 6 ]; in a footnote on page 133] that the occurrence of alkaptonuria, one of the "Inborn Errors of Metabolism" first reported by Garrod [ 10 - 12 ], was an example of Mendelian recessivity of a trait or character. Bateson, incidentally, coined the word "genetics". Another leading contender for the title "the first geneticist" was the Danish biologist, Johannsen [ 13 , 14 ], an equally enterprising experimentalist and astute thinker. Johannsen [ 14 ] was the first to define the term "das gen ; (plural) die gene" as the determinant of a trait; he was also the first to make a clear distinction between the genotype (der Anlagetypus) and the phenotype (der Erscheinungstypus ) on the basis of his experiments with self-fertilising bean plants. In Johannsen's experiments the weights of individual beans were the characteristics or traits. He had, in effect, repeated Mendel's experiments but by measuring a trait (individual bean weights in successive populations of plants) he was able to introduce three new concepts (gene, genotype and phenotype) that were the most significant, after Mendel's concepts of combination, segregation and recombination of traits during gametogenesis, in understanding the origin of genetic phenomena (the origin of changing traits). Failures to recognise the significance of Johannsen's work [ 13 , 14 ] prevented the development of rational concepts in genetics for at least the first two decades of the 20 th century. This failure is, surprisingly, still evident in current depictions of elementary Mendelian genetics (Section 3). 2.9. The tests of validity applied to Mendel's explanation for his observations It is clear that Mendel's experimental procedures (sections 2.2, 2.5) were sound; his notation was simple, unambiguous and consistently applied (section 2.6). His arguments (section 2.7) for a combination of traits in forming the F1 hybrids ( Aa ) are consistent with his arguments for the segregation of the component traits of the hybrid into separate gametes, and their random recombination in generating the F2 population ( A + 2 Aa + a ). Mendel's arguments pass the test of consistency. It is equally clear (but hitherto not noticed) that Mendel's explanations failed the test of plausibility. Mendel postulated that a F1 hybrid ( Aa ) was formed by combining the two differing traits ( A ) and ( a ) of their parents. He did not explain how a F1 hybrid ( Aa ) displayed only trait ( A ) and how it did not display trait ( a ), even when some F2 plants, like one of the two original parental (P) plants, did display trait ( a ). What explanation could we now give for this selective display of only one of two traits that are said to be combined in a hybrid? It may be (and has been) argued by some that trait ( A ) was displayed by the hybrids ( Aa ) because ( A ) was a dominant trait and ( a ) was a recessive trait. Such statements do not even qualify as a circular argument. They are illogical. Such statements fail to distinguish between an arbitrary definition and a plausible explanation. Mendel's definition of a dominant trait should be seen as an arbitrary device that accounts for his observation (by experiment) that his hybrids ( Aa ) in the F2 populations ( A + 2 Aa + a ) displayed trait ( A ) but not trait ( a ). A word of caution is necessary. Mendel's formulation ( Aa ) for a hybrid was crucial in establishing his consistent arguments; it was also the basis for Bateson's recognition that the essential features of Mendel's work were the concepts of combination, segregation and recombination of alternative traits (i.e., components of the phenotype). If we now wish to replace Mendel's implausible formulation ( Aa ) for a hybrid by a plausible formulation, we face the prospect of abandoning the rest of Mendel's arguments. That is not to say that we abandon admiration for Mendel's work. For its time, it was unsurpassed and should be recognised as one of the important steps in the development of experimental procedures in what became known as genetics. We should take care not to misrepresent Mendel's experiments and his arguments. It will become clear that misrepresentations of Mendel's paper have served only to sustain untenable concepts in current biology. In the post-Mendel era we assert that it is not components of the phenotype that segregate and recombine. It is the alleles (i.e., components of the genotype) that combine, segregate and recombine. May we then anticipate that modern explanations of Mendel's observations will pass the tests of consistency and plausibility? 3. Current accounts of elementary Mendelian genetics 3.1. Explanations of Mendel's observations The currently favoured explanation for Mendelian heredity in general, and in particular for the occurrence of Mendel's 3(dominant):1(recessive) trait ratio, is shown in Figure 2 . Figure 2 The currently favoured depiction of Mendelian inheritance following self-fertilisation of F1 hybrids represented by the allele pair ( Aa ). Section 3.1 of the text records the arguments commonly used in attempts to account for the alleged F2 trait series ( AA + 2 Aa + aa ) and for Mendel's 3(dominant):1(recessive) trait ratio. Sections 3.2 and 3.3 discuss the faults in these arguments. The assertions and descriptions generally attached to Figure 2 are as follows. (i) Mendel explained his experimental results by assuming that particles or factors (now called alleles) determined or specified the observed traits. (ii) ( A ) is a dominant allele; (a) is a recessive allele. (iii) The alleles in the male and female heterozygous somatic cells (Aa) segregate into separate gametes. Each gamete then contains only one dominant allele (A) or only one recessive allele (a). (iv) Fertilisation is a completely random event. Given a large number of fertilisation events, the possible recombinations of alleles are those displayed in the four squares. (v) Therefore the average distribution of the alleles at one diploid locus in the resulting progeny population of individual plants will be ( AA + 2 Aa + aa ). It is then argued that: (vi) The dominant allele pair ( AA ) will give rise to a dominant trait ( AA ). (vii) The recessive allele pair ( aa ) will give rise to a recessive trait ( aa ). (viii) In the heterozygote ( Aa ), the recessive allele is ineffective, or is suppressed by the dominant allele ( A ), so that only the dominant allele ( A ) is expressed in the heterozygote. Expression from one ( A ) is as effective as that from two dominant alleles ( AA ). Thus the heterozygote ( Aa ) expresses a dominant trait. (ix) Therefore the allele series ( AA + 2 Aa + aa ) is expressed (in a population of the progeny plants, animals or cells) as the trait series ( AA + 2 Aa + aa ). (x) This trait series gives rise to Mendel's 3(dominant):1(recessive) trait ratio (by the arguments in vi, vii, viii). 3.2. Faults in these currently favoured descriptions of Mendelian genetics There are seven faults in the descriptions and arguments attached to Figure 2 . (i) Mendel is misrepresented; he did not assume that particles or factors specified the observed traits. It is historically inaccurate and scientifically misleading to suppose that he made any such assumption. (ii) The letters ( A ) and ( a ) are Mendel's notation for dominant and recessive traits (Figure 1 , Table 1 ). If we are to continue to discuss Mendelian genetics, these notations (and the nomenclature dominant and recessive ) should refer to traits alone. (iii) Figure 2 fails to distinguish between the components of the genotype and the components of the phenotype (Johannsen, Section 2.8) because it asserts that alleles are dominant or recessive; and uses the same notation ( A and a ) and the same nomenclature ( dominant and recessive ) for both. (iv) Because we must not confuse alleles with traits, we could reasonably write an allele series as ( UU + 2 Uu + uu ); this states that a given locus, in three genetically related diploid cells, comprises a pair of two normal alleles ( UU ), or one normal and one mutant allele ( Uu ), or a pair of two mutant alleles ( uu ). Mutations change the allele constitution or composition at a locus. The modern (non-Mendelian) notation ( AA + 2 Aa + aa ) in Section 3.1 (items vi, vii) then states explicitly that a dominant trait ( AA ) comprises two aliquots ( A + A ) of some material substance or of two doses of dominance ( A + A ); likewise that a recessive trait ( aa ) is composed of two entitities ( a + a ) or two doses of recessivity. This is simply not true. It was not true in Mendel's time and it is not true today. Furthermore, it is not what Mendel's notation meant. It was pointed out (Sections 2.5, 2.6) that Mendel's notation ( A ) and ( a ) distinguished classes of traits, specifically 'constant form' classes of traits (Table 1 ). To substitute ( AA ) for ( A ) and ( aa ) for ( a ) in a trait series is illogical and indicates a regrettable failure to read Mendel's paper with the care that should be given to one of the classic papers in biology. (v) If the arguments attached to the homozygotes in Figure 2 are sound, they should also apply to the heterozygote. It is argued in Figure 2 that two dominant alleles ( AA ) generate a dominant trait ( AA ); and that two recessive alleles ( aa ) generate a recessive trait ( aa ). In other words, it is asserted that there is a direct, positive, linearly proportional (or additive) relationship between the allele constitution at a gene locus and the constitution of the trait expressed from that locus. If we are to be consistent, the same arguments should apply to the heterozygote ( Aa ). On the contrary, the arguments in section 3.1 (item viii) state that one dominant allele ( A ) in a heterozygote ( Aa ) is as effective as two dominant alleles ( AA ) in the homozygote. The arguments in item (viii) are therefore inconsistent with arguments in items (vi) and (vii). Item (viii) also transfers Mendel's implausible assertion that a hybrid ( Aa ) displays only trait ( A ) to the equally implausible assertion that one allele ( A ) in a heterozygote ( Aa ) is as good as two such alleles in a homozygote ( AA ). The argument in item (viii) that allele ( a ) is ineffective is an extreme case; it is therefore not generally applicable. The alternative argument, that allele ( a ) in a heterozygote is suppressed by the dominant allele ( A ), lacks any experimental support or rational theoretical justification. Items vi, vii and viii attached to Figure 2 are arbitrary, irrational and implausible devices applied to the heterozygote alone; they seem to have been introduced solely in order to arrive at the desired result. (vi) Figure 2 and the attached arguments thus fail to give rational explanations for the occurrence of dominant and recessive traits and for Mendel's 3(dominant):1(recessive) trait ratio. (vii) Figure 2 does not and cannot account for the observation that dominance and recessivity are not observed for all traits. The assertion in Figure 2 that the alleles are themselves "dominant" or "recessive" (and thus determine that traits are dominant or recessive) conflicts with inability of Figure 2 to explain why dominant and recessive traits are not always observed; nor does Figure 2 account for the observation that, when dominance and recessivity do occur, they do not always exhibit a 3:1 trait ratio. 3.3. Comments on these faults It is necessary to restate fault (iii) in section 3.2 in more widely applicable terms. It is illegitimate to use the same notation and nomenclature for a parameter and a variable in the same system. Parameters are those components of any system that are directly accessible to the experimentalist; they can be changed and maintained by the experimentalist at the new value, at least for the duration of an experiment. Variables are those components of the same system that are not directly accessible to the experimentalist; they can be changed and maintained at a new value only by making a finite change in at least one parameter of the system or of its immediate environment. The magnitudes of individual variables, in any system, respond to changes in the magnitude of one or more parameters of the system or of the immediate environment. In the case under discussion, the alleles are parameters (and part of the genotype); the traits are variables (and part of the phenotype). If the parameters and variables of any system of interacting components are represented by the same notation and the same nomenclature, confusion will inevitably result – as illustrated by Figure 2 , by the assertions (i) and (ii) and by the false arguments (vi) to (x). Traits may be dominant or recessive [ 1 ]; alleles cannot also be dominant or recessive. Figure 2 , and the arguments attached to it, fail all tests of consistency and plausibility (Section 2.9); they also fail the test of historical accuracy. 3.4. Another example of the improper transfer of dominance/recessivity from traits to alleles The primary error in Figure 2 is the illegitimate transfer of Mendel's terms "dominant" and "recessive" from traits (variables) to alleles (parameters), followed immediately by the reverse (and perverse) argument that the traits specified by the alleles must be dominant or recessive because the alleles are dominant or recessive . This habit is unscientific. It also occurs in discussion of mutations of non-catalytic proteins. When haemoglobin A (HbA) is mutated to the sickle cell haemoglobin (HbS), the three possible trait forms are correctly depicted as follows: (A/A) – the homologous, normal/normal protein, condition; (A/S) – the heterologous, normal/mutant protein (sickle cell), condition; (S/S) – the homologous, mutant/mutant protein, condition. Contrast these depictions with those sometimes found: (A/A) – the dominant condition; (A/S) – the sickle cell condition; (S/S) – the recessive condition. These latter statements depend solely on the illegitimate transfer of Mendel's terms dominant and recessive from traits (variables) to alleles (parameters) and the contention that, if alleles are themselves dominant or recessive, their expressed traits must always be dominant or recessive. If changes in the composition of non-catalytic proteins do explain the occurrence of Mendel's dominant and recessive traits, we require a demonstration that does not depend on these illogical notions. The sickle cell trait (A/S) in humans is significantly different from the normal trait (A/A). Those carrying the sickle cell (A/S) condition enjoy an advantage in areas where malaria is endemic . They do not die from malaria as frequently as those in the population with the (A/A) condition. The sickle cell condition (A/S) is debilitating but, provided it is not too debilitating, the frequency in the local population of those carrying the (A/S) protein pair is greater than it would be in malaria-free areas. This higher frequency of the sickle-cell (A/S) condition in areas where malaria is endemic is often said to be an example of "over-dominance". The term "over-dominance" is inappropriate. It presumably arose from the illegitimate transfer of the terms dominant and recessive noted above. The appropriate term is "heterozygous superiority". The "superiority" indicates the better chance of surviving in regions where malaria is endemic. 4. Conclusions: beginning a rational explanation for Mendel's observations The illegitimate use in Figure 2 of the same notation ( A and a ), and the same nomenclature (dominant and recessive), to describe an allele series and a trait series can be traced to Sutton [ 15 ]. Sutton asserted that the proportions of the chromosome pairs in the F2 population "would be expressed by the formula AA :2 Aa : aa which is the same as that given for any character in the Mendelian case." Mendel's expression ( A + 2 Aa + a ) gave the proportions of characters in his F2 population as A :2 Aa : a . Sutton gave no justification for rewriting these proportions in the form AA :2 Aa : aa . By writing the expression for chromosome pairs as AA :2 Aa : aa and the expression for the proportions of F2 characters as AA :2 Aa : aa , Sutton established a direct, one-for-one, relationship between pairs of chromosomes and the traits arising from them. This false relationship also persists in the currently favoured depiction of Mendelian genetics (Figure 2 ). Sutton's notation for pairs of chromosomes ( AA :2 Aa : aa ) was later transferred to pairs of alleles (what Sutton described as subunits of the chromosomes). It would be easy to blame Sutton for our present confusions. We should remember that Sutton, and those in the early years of the 20 th Century who copied his error, were struggling to understand the hereditary origin of traits. We may more reasonably ask: Why, one hundred years later, are these obvious errors still one of the features of Figure 2 ? Have these errors not been noticed before or, if they have been noticed, why they have not been corrected? Why also has the inconsistency and the implausibility of the arguments attached to Figure 2 not been noticed or corrected? Why (in both of the examples given in sections 3.3 and 3.4) are alleles (components of the genotype) not distinguished, as they surely should be in genetics, from traits (components of the phenotype) by using different notations and nomenclatures for each? Traits (variables) may be dominant or recessive, as defined by Mendel. Alleles (parameters) are, always have been, and can only be normal or abnormal (mutant). Harris (pages 143–157 in reference [ 16 ]), for example, referred consistently to normal and abnormal alleles (not to dominant and recessive alleles), whereas, as noted above, alkaptonuria was a Mendelian recessive trait or character (page 133 in reference [ 6 ]; page 19 in reference [ 16 ]). A review of 13 textbooks of genetics showed that in 12 instances, dominance and recessivity were defined specifically as properties of genes or alleles. These texts, published between 1982 and 2002, were intended for student use; their definitions of dominance and recessivity ignore Mendel's definition of dominance and recessivity as properties of the traits (sections 2.4, 2.5, 2.6, 2.7); they take no account of the need to distinguish between the parameters and variables of a system of interacting components (section 3.2). In one of these 12 texts, it was further claimed that: "Mendel proposed the existence of what he called particulate unit factors for each trait". In another, that: "Mendel realised that some genes (dominant genes) expressed themselves when present in only one copy". In a third that: "Mendel imagined that during the formation of pollen and egg cells, the two copies of each gene in parents segregate". Of these three quoted texts: The first misrepresented Mendel; he did not "propose the existence of particulate unit factors for each trait". The second misrepresented Mendel by transferring his term dominirende ("dominating") from traits to genes; the second and the third quoted texts ignored the fact that the term " das gen " (plural " die gene ") was first used and its role as the determinant of traits postulated by Johannsen, 43 years after Mendel's paper was published (Section 2.8); Mendel did not mention the word gene (Section 3.2). Of the 13 texts examined, only one gave a definition of dominance and recessivity that would have been recognised by Mendel. Even so, this author contradicted his correct definition of dominance and recessivity as properties of components of the phenotype by giving an explanation of elementary Mendelian genetics that employed Figure 2 and its associated arguments. All 13 of the texts examined ignored or contradicted the verifiable historical evidence (sections 2.2–2.7) and failed to make the obligatory distinction between the functions of alleles and the properties of traits. The correct nomenclature for alleles used by Harris (pages 143–157 in reference [ 16 ]) is, unfortunately, rarely if ever employed by other authors. Pasternak [ 17 ], for example, accepted that "in strict genetic terms, dominance and recessivity are descriptions of the phenotype and not of the genes." but then continued: "However, few textbooks bother to make the distinction, because it was both convenient and highly ingrained for geneticists and others to refer to dominant and recessive alleles." Ingrained it may be, convenient (and scientifically legitimate) it is not. If we continue to propose Figure 2 and the attached arguments as an explanation of Mendel's work, we deceive ourselves and encourage irrational thinking in our students at a time in their education when they are most vulnerable. It is extraordinary that an "explanation", like Figure 2 , should still be found in textbooks intended for student instruction; it exposes our own confusion but explains nothing of scientific value in genetics. Any student who criticised Figure 2 and the attached arguments in an answer to an examination question would have shown commendable scientific insight but, according to current teaching, would be deemed to have failed that question. Barker [ 18 ], writing on another topic, suggested that it might take 50 rather than 25 years for textbooks "to get it right". On the evidence presented here, Barker was too optimistic. The four errors introduced by Sutton [ 15 ] remain uncorrected (Figure 2 ) 100 years later. To be fair to authors of textbooks of genetics, every author inevitably relies on what has been written by preceding authors. However that may be, we are faced with an uncomfortable question. Are we content to continue to deceive ourselves, to give our students a false picture of what Mendel achieved, and to provide them with untenable 'explanation' of his remarkable observations (Figure 2 )? Presumably not, especially when we can very easily begin, in this article, to construct a rational explanation for Mendel's observations and for other observations of current interest in genetics. A fresh approach to the origins of dominant and recessive traits is needed. As a first step, we need to represent normal and mutant alleles by symbols that differ from those ( A, a, B, b, C, c ) used by Mendel to represent traits . We must replace symbols ( A and a ) for alleles in Figure 2 by quite different symbols; e.g. ( U ) to represent a normal allele, not a "dominant allele"; and ( u ) to represent a mutant or abnormal allele, not a "recessive allele" . The F2 allele series in Figure 2 would then be, on average, UU + 2 Uu + uu. Similarly, the trait series in Figure 2 must be replaced by Mendel's notation (A + 2 Aa + a) because, as explained earlier, Mendel was concerned (as we are, first and foremost) only with understanding the origin of two classes of trait – the dominant class ( A ) and recessive class ( a ). We will later be concerned with the quantitative composition of traits. We have, however, already identified an implausibility in Mendel's notation ( Aa ) for a hybrid that, allegedly, displayed the trait ( A ). An implausibility, like an inconsistency, must be eliminated if we are to arrive at an internally consistent and plausible account of Mendel's observations. The implausible notation ( Aa ) can be eliminated by replacing it by the single symbol ( H ) for a hybrid. We have now adopted a stance that, in sharp contrast to Figure 2 , distinguishes clearly between determinants and that which is determined. We have allocated a nomenclature and notation for alleles that is distinct from that allocated to traits. We have differentiated clearly between the parameters of the system (in this particular case, the components of the genotype) and the variables of the system (in this particular case, the components of the phenotype). Mendel found, by experiment, that the proportions of different plant forms in his F2 populations were 1(dominant trait):2(hybrids):1(recessive trait) or, in his notation, ( A + 2 Aa + a ). Replacing Mendel's notation ( Aa ) for a hybrid by the single symbol ( H ) does not alter Mendel's experimental observation of the proportions of trait forms in the F2 populations (section 2.5). It does mean that we can avoid Mendel's implausible postulate that, although recessive trait plants did display trait ( a ), his hybrids ( Aa ) did not. We have, of course, to discover an experimentally verifiable mechanism that would explain why hybrids ( H ) display a trait that is sometimes indistinguishable and sometimes distinguishable from trait ( A ). Our remaining task is to explain rationally how this series of normal and mutant alleles (UU + 2 Uu + uu) in the F2 population is expressed as the trait classes ( A + 2 H + a ) in that population, where all that we have done is to replace Mendel's implausible ( Aa ) by a plausible ( H ). Note also that we have now also eliminated the illegitimate use of paired symbols for Mendel's dominant ( A ) and recessive ( a ) traits. Most of the clues that facilitate this task are present in this article. One clue is missing, but it can be inferred by asking how one allegedly dominant allele ( U ) in a heterozygote ( Uu ) could be as effective as two such alleles ( UU ) in a homozygote. A further article will provide the answers, but in the interval readers may like to rise to the challenge of explaining: (1) how dominant and recessive traits arise from normal and mutant alleles , and (2) why Mendel's 3:1 trait ratio, though not uncommon, does not always occur.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC516238.xml
554989	Automated migration analysis based on cell texture: method & reliability	Background In this paper, we present and validate a way to measure automatically the extent of cell migration based on automated examination of a series of digital photographs. It was designed specifically to identify the impact of Second Hand Smoke (SHS) on endothelial cell migration but has broader applications. The analysis has two stages: (1) preprocessing of image texture, and (2) migration analysis. Results The output is a graphic overlay that indicates the front lines of cell migration superimposed on each original image, with automated reporting of the distance traversed vs. time. Expert preference compares to manual placement of leading edge shows complete equivalence of automated vs. manual leading edge definition for cell migration measurement. Conclusion Our method is indistinguishable from careful manual determinations of cell front lines, with the advantages of full automation, objectivity, and speed.	Background The images we deal with are images of endothelial cells. Endothelial cells migrate to sites of injury in the body. They are involved in forming new blood vessels to help repair damaged areas [ 1 , 2 ]. In particular, we are interested in the effects of SHS on endothelial migration. By comparing automated migration analysis with varied exposure to SHS, for cells with and without specific genes, we can examine why exposure to SHS impairs endothelial cell migration and explore possible cures [ 3 , 4 ]. Cell migration is a basic biologic function that can be modified by changes in genetic code and in response to chemical and other stimuli. Upon 24 hours serum starvation, the cells were artificially wounded using P20 pipette tip across the plate, then cultured respectively in regular DMEM or DMEM containing Second Hand Smoke (SHS) (unpublished). The subsequent gaps were imaged at 0 and 6 hours post SHS exposure as previously described to determine the rate of migration of the front lines to close in the gap [ 5 - 8 , 23 ]. Figure 1 gives an example about the cell migration. Figure 1 Images of a plate of endothelial cells growing in agar. Cells appear as dark spots and contrast is limited because introduction of stains to increase contrast could affect cell function. An early time image (top row) shows a wide trough where a lane of cells was removed by the experimenter. A later time points (bottom row) show the effect of progressive narrowing of this lane as the cells migrate to fill in the gap. Our goal is to measure the width of the "clear lane" which corresponds to the amount of cell migration in the time interval. The automated borders are compared blindly by a team of domain experts to manual borders created by a technician to assess accuracy. Results are also evaluated blind to biologic significance to determine concordance and power to demonstrated biological effects. Biologists deal with this by making multiple manual measurements, to report an average. Observers have difficulty deciding where and how many times to measure the width. Besides, there are many pairs of images to be processed. Therefore, automatic measurement is desired. Implementation Generating texture The primary difference between cell-populated areas and the clear lane is texture. The cell-populated areas are speckled with cells, the clear lane is not. In order to capture the cellularity characteristic of the source images, we sought to compute a texture index that would emphasize the cellular attribute of the region of interest and also minimize the influence of non-cellular signal variations [ 11 - 14 ]. Because the image may have non-uniform background where the "clear lane" can be 'darker' than the 'cells' at other locations, the texture index should be generated from the relative gray value difference. Furthermore, we know the orientation of the experimentally produced clear lane, which we take to be vertical. Then our algorithm generates the texture in this way: For each point in the original image Search for darker point in this line vertically If find Set the distance between start and darker point as the gray value of the corresponding point in the texture map Search for continual darker points and set the distance as the value of them Scale the value to 0–255: where pv is the new pixel intensity value, cv is the distance value of the corresponding point and max and min are the minimum and maximum distance value. Panel b of figure 2 shows the texture of Panel a of figure 2 using this algorithm. Figure 2 An example image is shown in panel (a), followed by the derived texture index (b) and the resultant graphic overlay (c & d). Note that the region between the vertical lines of 2D is relatively devoid of cells, and each line represents the front of cell migration, as desired. Migration analysis Based on the texture map, the region we are interested in appears as a white vertical band. Thus the second stage of analysis must determine the position and width of this lane. As the information in each vertical column is equivalent to repeated measures, we can combine the data to marginal projection. From the histogram of this we can compute a classifier for lane vs. cells and determine the half-height width. The locations are then mapped to a graphic overlay on the original image to demarcate the front lines of cell migration. The change in distance between the front lines reports the amount of (bi-front) cell migration. 1. Project the texture values to a marginal profile "cellularity index profile" (Figure 3 ) which is an array of P --- P [1.. n ] where n is the width of the image. The value of each P [ i ] is the intensity sum of all the pixels in i column. Figure 3 The marginal cellularity index profile. 2. Compute the discriminant classifier (DC) which is average value of P [ i ]. 3. Locate the leading and trailing edges based on classifier crossing. If P [ i ] < DC and P [ i +1] > P [ i ], i is the leading edge. Conversely, if P [ i ] > DC and P [ i +1] < DC, i is the trailing edge. Then a few pairs of leading and trailing edges could be obtained. The target pair is identified based on the width, P [ i ] values between leading and trailing edges and "clear lane" location in the time neighboring image. 4. Record locations and generate graphic overlay for original image. Validation Since the manual assessment is the research gold standard for image processing [ 15 - 18 ], a technologist specially trained to identify the leading and trailing edges of cell migration was provided a computer tool to mark those edges manually in a manner compatible with the graphics overlay engine. These are called "manual edges." The manual edges and the automated edges were then presented to a team of domain experts in random order pairs (one of each on corresponding image) for preference scoring. The scores ranged from 1–5, where 1 is strong preference for first overlay, 2 mild preference for first overlay, 3 equivalency, 4 mild preference for second overlay, and 5 strong preference for second overlay. Results were analyzed by Kappa statistic as a measure of agreement. Results After analysis, results like the ones shown in the Figure 2 are obtained. Panel (a) shows a photograph of the cell cultures, while the remaining panels show various aspects of the analysis. Figure 4 shows the two worst cases of disagreement between automated and manual methods. Figure 4 Examination of width differences between automated and manual identified two outliers with relatively large differences. These are shown as two pairs. Note the uneven heterogeneous cell distribution in these cases, a result that is technical suboptimal and not desired, likely from an error in the excoriation (creation of clear channel). These are a poor sample pairs for technical reasons, and in retrospect domain expert still had no significant difference in preference of manual result over the automatic result. The results of domain expert preference by quality for automated vs. manual assignation of migration front lines, evaluated blinded to method, randomized, and subsequently decoded. Overall, there is complete equivalence of automated vs. manual with respect to expert preference for quality. The values ranged 2–4. In no cases was manual strongly preferred over automatic. Preference testing of analysis methods showed near equivalence, favoring preference for the automated borders (3.02 ± 0.11). Agreement between observers in preference was examined for two domain experts, revealing good agreement (Kappa = 0.59, p < 0.003). Agreement in preferences by a technologist without domain expertise was lower (Kappa = 0.23, 0.25, p > 0.10) but supported the same conclusion: the automated analysis is at least as good as manual selection by domain experts. Application of this method to determine the effect of SHS on endothelial cell migration demonstrates that SHS can reduce the cell migration rate, which is statistically significant (Figure 5 ). Figure 5 The migration distance of control group (NL) is 101.30 ± 10.32 and distance of SHS group 36.25 ± 2.71. Pair t test shows two groups are obviously different with p value = 2.24E-06. Discussion Our migration analysis is based on the texture index of the images. This index should reflect the attribute of the images. Since no global thresholding technique could be used in our images, the segmentation of regions and boundaries (edges) have to consider the local property [ 9 ]. Because the target boundaries always show as a vertical band, the line-based segmentation appears to be the most suitable approach for our task. Further analysis of regions and edges is based on a uniform data structure reflecting the texture character in each column. Our results show a robust automatic method with no detected errors. This study is a pilot study demonstrating feasibility and biologic significance in real application. Further collective experience in multi-center applications are needed to establish the full utility of the method. In addition, the program runs on the software platform, ImageJ [ 10 ] and the speed is fast. A normal process time for one study of images is less than 3 minutes. Results such as width and percentage can be shown as a table. It offers a convenient way for researcher to process their image data using excel. Conclusion We describe a novel method of cell migration analysis based on texture pre-processing and discriminant analysis. Domain expert preference testing demonstrates that this automated method compares favorably to the much more painstaking manual method. The further study is to apply this to evaluation of the impact of SHS on endothelial cell migration. For that purpose, we have constructed a SHS capture system in which we bubble the SHS through tissue culture medium to assess its impact on cell migration. Our results indicate that this analysis system is very sensitive to biological effects, documenting that SHS impairs cell migration [ 19 - 22 ]. Availability and requirements Project name: Cell migration measurement project Project home page: Operating system(s): Platform independent Programming language: Java Other requirements: Java 1.3.1 or higher, ImageJ License: Null Any restrictions to use by non-academics: Licence needed Authors' contributions JDP proposed and designed the method to evaluate cell migration objectively and automatically. JQ implemented the method as a plugin of ImageJ and validated the method. TWC performed the cell experiments and captured the images. LG performed manual edge definition and subsequently performed statistic analysis of expert preferences. All authors read and approved the final manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC554989.xml
551601	Two-part permutation tests for DNA methylation and microarray data	Background One important application of microarray experiments is to identify differentially expressed genes. Often, small and negative expression levels were clipped-off to be equal to an arbitrarily chosen cutoff value before a statistical test is carried out. Then, there are two types of data: truncated values and original observations. The truncated values are not just another point on the continuum of possible values and, therefore, it is appropriate to combine two statistical tests in a two-part model rather than using standard statistical methods. A similar situation occurs when DNA methylation data are investigated. In that case, there are null values (undetectable methylation) and observed positive values. For these data, we propose a two-part permutation test. Results The proposed permutation test leads to smaller p -values in comparison to the original two-part test. We found this for both DNA methylation data and microarray data. With a simulation study we confirmed this result and could show that the two-part permutation test is, on average, more powerful. The new test also reduces, without any loss of power, to a standard test when there are no null or truncated values. Conclusion The two-part permutation test can be used in routine analyses since it reduces to a standard test when there are positive values only. Further advantages of the new test are that it opens the possibility to use other test statistics to construct the two-part test and that it avoids the use of any asymptotic distribution. The latter advantage is particularly important for the analysis of microarrays since sample sizes are usually small.	Background The addition of a methyl group at the carbon-5 position of cytosine is a modification of DNA called DNA methylation. In mammalian cells, DNA methylation is essential for proper development [ 1 ]. The methylation patterns of tumor cells are altered compared to those of normal cells, moreover, there are also differences between different types of cancer as shown for subtypes of leukemia [ 2 ] and lung cancer [ 3 ]. Thus, DNA methylation analysis promises to become a powerful tool in cancer diagnosis [ 4 ]. DNA methylation data can be obtained using the MethyLight technology [ 5 ]. When the tested region is not or only partially methylated the result is negative (undetectable methylation, null values). In contrast, samples that show methylation will have a value greater than 0 [ 4 ]. Thus, DNA methylation data obtained with MethyLight have a clump of zero observations and a continuous nonzero part. For such a data structure, two-part models as proposed by Lachenbruch [ 6 - 8 ] are applicable. In that approach, the test statistic is the sum of two squared statistics, one comparing the proportions of zeros and one comparing the positive values. For example, one can use the binomial test and the Wilcoxon rank sum test. The asymptotic null distribution of the sum of the squares of the two test statistics is χ 2 with two degrees of freedom (df = 2). In microarray data it is relatively common that small and negative expression levels were clipped-off to be equal to an arbitrarily chosen cutoff value. Recent examples used different cutoff values: 1, 20, and 50, respectively [ 9 - 11 ]. Aside from the fact that negative values make no biological sense, there are, with regard to oligonucleotide arrays from Affymetrix, two primary reasons for truncating the values [ 12 ]. First, spots at the low intensity range are generally more vulnerable to noise, thus, it is thought that the technology produces a poor discrimination at low levels of expression [ 13 ]. Second, the focus is on expression of identified genes and expressed sequence tags. Differences at negative or low values may result from differences in binding to the mismatch probes. Since it is generally not known what binds to the mismatch probes, the differences at negative or low values cannot be attributed to target genes. Due to the truncation there are two different types of data: truncated values and original observations. Since the truncated values are not just another point on the continuum of possible values, it would be inappropriate to use a standard statistical method that would treat all values equally [ 14 ]. The two different types of data should be analyzed separately. Therefore, the two-part model, comparing the proportions of truncated values and the distribution of positive values, is applicable. Note that negative expression levels are not possible when the Affymetrix Microarray Suite (MAS) 5.0 software is used. However, small values are possible and may be truncated. Since, in microarray experiments, the sample sizes, i.e. the numbers of replications, are usually very small [ 15 , 16 ], the use of the asymptotic distribution of Lachenbruch's two-part test statistic may be questionable. Thus, we carried out permutation tests with the two-part statistic. For a permutation test all possible permutations under the null hypothesis are generated. In our situation we permute the group labels for the whole sample, i.e. for truncated values (or the null values in case of methylation data) and original observations. Then, the test statistic is calculated for each permutation. The null hypothesis can then be accepted or rejected using the permutation distribution of the test statistic, the p -value being the probability of the permutations giving a value of the test statistic as supportive or more supportive of the alternative than the observed value [ 17 , 18 ]. Thus, inference is based upon how extreme the observed test statistic is relative to other values that could have been obtained under the null hypothesis. We found that, in the case of a two-part model, the permutation test is not only a way to avoid the use of an asymptotic distribution, but also is a more powerful test, i.e. a test that produces, on average, smaller p -values. In addition, the permutation test reduces, without any loss of power, to a single test if no truncated (or null) values were present. Thus, the proposed test is applicable in routine use whether or not truncated (or null) values occur. After the definition of the tests in the following section, we present our findings for DNA methylation data and microarray data. We then confirm the results using simulations. Two-part tests As briefly mentioned above a two-part test statistic is the sum of two squared statistics, one comparing the proportions of truncated values and one comparing the positive values. Let n 1 and n 2 be the numbers of independent observations regarding one gene (or one region in case of methylation data, respectively), for two groups to be compared. The observed numbers of truncated values (or null values in case of methylation data) in the two groups are denoted by m 1 and m 2 . To compare these numbers m 1 and m 2 Lachenbruch [ 6 ] used the statistic where , , and . Under the null hypothesis the proportions of truncated values are not different between the two groups, and B 2 is asymptotically χ 2 -distributed with df = 1. B 2 is always well defined, unless there are only truncated values in both groups or no truncated values at all. For these two extreme cases we set B 2 = 0. For the second part in Lachenbruch's two-part model one can use different tests, Lachenbruch [ 6 - 8 ] considered the Wilcoxon rank sum test, Student's t test, and the Kolmogorov-Smirnov test. The use of the latter test in a two-part model, however, was too liberal, the type I error rate was close to 0.065 (for a significance level of α = 0.05 [ 7 , 8 ]). Both other tests work well. Here, we apply the Wilcoxon test for two reasons. This test was used in the original analyses of the data we use below [ 3 , 11 ]. Nonparametric tests based on ranks are more appropriate for non-normally distributed data such as microarray data [ 19 ]. The standardized rank sum statistic based on the non-truncated (or positive values in case of methylation data) values is defined as where RS is the rank sum, i.e. the sum of the ranks in group 1. When there are ties within the non-truncated values the denominator slightly changes [[ 20 ], p. 109]. For the extreme case that there are only truncated values in at least one group we set W = 0. The test statistic for the two-part test is X 2 = B 2 + W 2 . Under the null hypothesis of no difference between the groups, X 2 is asymptotically χ 2 -distributed with df = 2 [ 6 , 7 ]. Alternatively, a permutation test can be performed with the statistic X 2 . This test, called two-part permutation test here, is a permutation test based on the sum statistic X 2 . It is carried out by permuting the group labels for the whole sample. Thus, all observations, truncated and non-truncated values (or null and positive values in case of methylation data) are reallocated to the groups. When performing this two-part permutation test the exact permutation distribution of X 2 is determined. This distribution, computed by generating all possible permutations or, for the p -values given below in Table 1 , using a simple random sample of 20,000 permutations, is used to compute the p -value. Since it is a permutation test based on the sum X 2 , it is neither necessary to determine the permutation distributions of the summands B 2 and W 2 nor to calculate the p -values of the univariate tests related to B 2 and W 2 . Application to actual methylation data We use DNA methylation data from 7 regions and 87 lung cancer cell lines, 41 lines are from small cell lung cancer and 46 lines from non-small cell lung cancer [ 3 , 4 ]. The proportion of positive values for the different regions ranges from 39 to 100% for the small cell lung cancer and from 65 to 98% for the non-small cell lung cancer. The data are available at . Siegmund et al. [ 4 ] transformed the data by standardizing the positive values on the natural-log scale. However, for the tests applied here, this transformation has no influence. Table 1 presents the p -values of the two tests. For most regions the two-part permutation test gives a smaller p -value than the original two-part test. The only exception is the region APC . However, the original two-part test's p -value for this region is 0.1684 and, for a p -value of this size, a small change in the value is usually of no importance. Application to actual microarray data Tschentscher et al. [ 11 ] performed an experiment with HG-U95Av2 oligonucleotide arrays in order to compare patients with uveal melanomas with and without monosomy 3. Expression values were calculated by use of the MAS 4.0 software. The data are available at . The sample size in this microarray experiment is 10 per group. As mentioned above, expression levels below 50 were set to 50. This data truncation occurred in 2,215 (28%) out of 7,902 genes. First, we consider these 2,215 genes. Figure 1 displays the number of genes for the different number of truncated values per gene. Table 2 shows the frequencies of different size groups of the p -values. Often, the two-part permutation test gives a smaller p -value than the original two-part test. For instance, for 19 genes the p -value is ≤ 0.001 when the latter test is applied. The permutation test gives a p -value ≤ 0.001 for these 19 and for 39 additional genes. As usual, the majority of genes do not show any indication of being differentially expressed. For these genes a slight change in the p -value is of no importance. Thus, in Table 3 , we consider the genes for which the p -value of the original two-part test is ≤ 0.1. Out of the 2,215 genes 514 remain. As shown in Table 3 the p -values of the two-part permutation test are, on average, distinctly smaller than those of the original two-part test. For a large proportion of genes (72% in this data set) there are no truncated values. In that case, the two-part statistic reduces to the sum 0 plus the squared standardized rank sum W 2 . Of course, one has to define a priori whether the two-part test or the Wilcoxon test will be used to analyze the data. If the original two-sided (asymptotic) Wilcoxon test were chosen and applied, one could compare W 2 with critical values from a χ 2 distribution with one degree of freedom (df). If the two-part were chosen there is df = 2 and, in case of no truncated values, power is lost compared to the original Wilcoxon test. For instance, the 95% percentile of the χ 2 distribution with df = 1 is 3.84, but it is 5.99 for df = 2. The permutation test using the sum statistic X 2 does not suffer from this power loss: When there are no truncated values there is, of course, no difference in the proportions of truncated values, and the test statistic X 2 is, for every permutation, the sum 0 + W 2 . Thus, the two-part permutation test reduces to the exact two-sided Wilcoxon rank sum test when there are no truncated values. Consequently, this permutation test does not only give smaller p -values, but it is also applicable in routine use whether or not truncated values are present. Simulation study The two different tests, the original two-part test and the two-part permutation test, were compared in a Monte Carlo simulation study performed using SAS version 8.2, 5,000 simulation runs were generated for each configuration. The sample size of 10 per group was chosen as in the microarray experiment presented above. In some configurations some randomly chosen values were set to 0 according to binomial distributions with the probabilities p 1 and p 2 , and the remaining observations were generated according to a lognormal distribution (with median 1 and σ = 1). Then, the values of one group were shifted if applicable. In some other configurations all observations were generated according to the lognormal distribution and, in one group, shifted. Then, values smaller than a cutoff value were truncated. The type I error rates of the two tests are very similar. With e.g. p 1 = p 2 = 0.3 and a significance level of α = 0.05 the simulated type I error rates were 0.049 for both the original two-part test and the two-part permutation test. Table 4 displays results for situations with a difference between the two groups. As above, only those comparisons were regarded for which there is some indication of a difference, i.e. a p -value ≤ 0.1 of the original two-part test. In all considered configurations the median of the difference between the p -values of the original two-part test and the two-part permutation test is positive. The finding that the p -values of the permutation test are smaller corresponds to a higher power of this test. The power is given in Table 5 , as shown the power of the two-part permutation test is at least as high as that of the original two-part test. There is only one exception, the latter test is slightly more powerful in one situation, i.e. when the proportion of zeros is higher in group 1 and the positive values are larger in group 1. Discussion Previous research demonstrated that a two-part test is appropriate and powerful in the presence of a clump of zero observations (i.e. truncated or null values). In this paper we propose a permutation test for such situations with two types of data. Usually, nonparametric tests can be performed based on an asymptotic distribution or based on a permutation null distribution. The two approaches often give similar results, especially when the sample sizes are large. However, in the case of a two-part test one cannot simply replace the asymptotic distributions of B 2 and W 2 by the exact permutation distributions. If so, one would compute two exact p -values although the aim of a two-part test is to receive one p -value that combines information from both parts. Therefore, the two-part permutation test uses the exact permutation distribution of the sum statistic X 2 . That this permutation distribution of the sum is generated rather than to simply replace the asymptotic distributions of the summands by their exact permutation distributions may be the reason why the permutation test is more powerful. A disadvantage of a permutation test is that it can be computer-intensive. However, this issue is less relevant now due to faster algorithms [[ 18 ], chap. 13] and the advent of high-speed PCs. Furthermore, one can carry out a permutation test based on a random sample out of the possible permutations, as we did for the DNA methylation data (see Table 1 ). In microarray experiments it is common to investigate thousands of genes simultaneously. The approach presented here for the identification of differentially expressed genes is to consider a univariate testing problem for each gene. A correction for the multiplicity of genes is a subsequent step, that is, like the previous step of normalizing the data, outside the scope of this paper. A common approach to the multiplicity problem is to consider a procedure for testing the genes simultaneously for differential expression with the test on an individual gene being implied in the simultaneous test. For such a procedure different proposals have been made recently. For instance, there are methods based on the p -values of the tests from individual genes [ 21 - 23 ]. In a similar manner, the multiplicity of regions can be managed in DNA methylation data. Conclusion Aside from the shown improvement in power, the proposed two-part permutation test has three important advantages. First, it avoids the use of any asymptotic distribution and, therefore, can safely be applied in case of small sample sizes that are common in microarray experiments. Second, it reduces without any loss of power to the exact Wilcoxon test if there were no truncated (or zero) values. Thus, it can be used in routine analyses. Third, the permutation test opens the possibility to use other tests to construct the two-part test. Thus, tests with unknown or non-standard null distributions can be used. For instance, one could replace the Wilcoxon test by the Baumgartner-Weiß-Schindler test [ 24 ] that was recently recommended for the analysis of gene expression data [ 19 ]. Authors' contributions MN performed the statistical analyses and drafted the manuscript. TB prepared the microarray data. TB and KHJ participated in the design of the simulation study and helped to draft the manuscript. All authors read and approved the final manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC551601.xml
552334	Anal screening cytology	This issue of CytoJournal contains an article on screening for anal intraepithelial neoplasia in high-risk male patients. This accompanying Editorial focuses on current understanding of this relatively new disease entity, with insights as to the potential role of screening cytopathology in the epidemiology, pathophysiology and clinical management of this HIV and HPV related anal lesion, which predominates in male patients living long-term with AIDS. Mention is made of techniques of obtaining samples, methods of preparation, and morphologic classification. Issues of anoscopic confirmation, as well as topical and surgical management are emphasized. The similarity of initial experiences in anal screening to problems encountered early in cervical cancer screening programs several decades ago, are highlighted.	Note For corresponding research article please see Arian et al, 2005 [ 15 ] The subject of anal screening cytology has entered the epidemiologic and cytopathologic literature as a topic of interest over the last decade, and is highlighted in this issue of CytoJournal in an article by Arain and colleagues. Until recently, anal cancer was not considered to be a neoplasm of major public health concern [ 1 ]. It occurred infrequently, usually in older people, affecting women more often than men; further, being a neoplasm of low incidence, it remained under the radar in terms of screening potential. In some respects, anal cancer mirrored cervical cancer in unscreened women, presenting late in the course of the disease with a variety of pelvic symptoms, and having a protracted, ultimately fatal, course. Patients invariably became social outcasts, suffering from intolerable fecal complications, which presented major nursing challenges. Two aspects of this scenario have changed. First was the fairly recent introduction of effective modern treatment regimes for invasive anal squamous cancers utilizing chemo-radiation, leading to improvement in morbidity and long term survival [ 2 , 3 ]. The second significant alteration was seen in epidemiology, and this is the area which has come to involve screening cytopathology. During the 1990s, in several European and North American cancer centers, an initially unaccountable increase in anal cancers was seen in younger people. It soon became apparent that HIV-positive homosexual males were affected in numbers greatly in excess of those expected. This was first noted in those urban areas in which large concentrations of homosexual men had been treated since the outbreak of the HIV epidemic. The term "males having sex with males" (MSM) was coined to cover these high-risk individuals. Preliminary information gleaned from screening programs in high HIV-positive incidence areas amongst MSM, showed detection rates of intraepithelial and early invasive neoplasia higher than any incidence ever recorded for cervical cancer screening. Further, HIV-negative homosexual MSMs, and HIV positive non-homosexual men (eg drug users) also exhibited an increased incidence of pre-malignant and invasive anal carcinomas. More recently, the syndrome of early onset anal cancer has been extended to include HIV-positive female patients, as well as females who are not HIV-positive, but who have genital HPV. All these groups show a higher incidence of abnormal anal cytopathology, though none quite as high as that found in HIV positive MSM. The unifying factor in most instances is ano-receptive intercourse, or extension of HPV infection from the genital tract to the anal mucosa, most obvious in the face of deficient immunity, or high viral load [ 4 - 6 ]. Epidemiologic studies have indicated that a prolonged preclinical phase precedes the onset of anal cancer in these high-risk groups. Just as in the cervix, there is a transitional zone (although not an abrupt squamocolumnar junction) in the anus. Rectal mucosa, with goblet cells, ends about one inch proximal to the external sphincter, giving way to a transitional epithelium, which in turn blends into a stratified squamous epithelium at the level of the anus. (External to the anus, any lesions which arise are considered to be primary skin lesions rather than anal lesions.) Almost all anal cancers develop in the transitional zone, where atypical, dysplastic and in-situ lesions are identifiable histologically. The cytologic counterparts are those of atypical cells of uncertain significance (ASCUS), low grade squamous intraepithelial lesions (LSIL) and high grade squamous intraepithelial lesions (HSIL). Once this natural history had been demonstrated and confirmed, it seemed natural that exfoliative cytology could be investigated as an "anal Pap smear" [ 7 , 8 ]. As with the cervix, the technique of obtaining the anal sample is critical to the success of screening. Standard colonic preparation is not required, but the rectum should be emptied prior to obtaining the anal sample. Brushes, brooms and Dacron swabs have all been used, the type of spatula probably being less important than the skill of the operator. The instrument is inserted to a depth of one and a half inches beyond the external sphincter, and subsequently withdrawn in a firm downward spiral movement incorporating 10–12 rotations, to ensure the device has made contact with the full surface area of the transformation zone. Some have used direct smears onto glass slides with immediate wet fixation; most centers, however, employ immediate insertion of the scraping device into liquid fixative for thin layer preparation. This appears both to improve adequacy and preservation, and also to eliminate any fecal contamination; residual material in the vial can be used for HPV studies if required, or for the creation of a bank of teaching slides. There has been a relative dearth of literature on the cytomorphology of anal samples [ 9 - 11 ]. Classification according to Bethesda guidelines has been advised, implying similarity of the exfoliative cytology of atypical squamous cells of undetermined significance (ASC-US), anal intraepithelial lesions (A-SIL) and invasive squamous-cell neoplasms to those encountered in the cervix. This early in the development of the science of anal screening, and evident also in the current article in CytoJournal, there are indications that cytologic evaluation may not fully anticipate the severity of some lesions, when compared with biopsies taken simultaneously. Several groups have advised that all abnormal anal Paps should be followed by anoscopic evaluation, with biopsy confirmation when necessary. Anoscopy essentially mirrors colposcopy, enabling the viewer to see vascular abnormalities at magnification, and direct biopsies to the most abnormal-appearing areas. Despite these basic similarities, it is strongly stated in the literature that expertise in colposcopy does not equate to immediate competence in anoscopy; a significant learning curve exists for those wishing to acquire excellence in anoscopic technique, with associated accurate biopsy sampling. It is important that the anoscopist not be sidelined by visible condylomata, which may merely be sentinels of deeper flat lesions of higher grade. At the present time, lack of available expertise in this interventional follow-up of abnormal anal Paps may be the single limiting factor in any new screening program. It thus behooves those interpreting anal samples to be as proficient as they can be in pre-interventional assessment of the anal transformation zone. Experience over the last decade suggests that anal cancer may be as highly appropriate a target for screening as is the cervix, in selected populations. The neoplasm is frequently encountered in well-defined high-risk groups. It has a detectable pre-malignant phase, and is amenable to easy cytologic sampling. Cytodiagnosis is reasonably sensitive and highly specific, and histologic confirmation is relatively easily obtained by well-trained personnel. If there is a current area of deficiency in such programs, it may well be in the treatment of intraepithelial lesions, which, as yet, has not been adequately assessed in large numbers of patients. It is known that highly active anti-retroviral therapy (HAART) does not appear to alter the pathophysiology of anal lesions once initiated. A variety of topical agents such as podophyllotoxin and imiquimod have been tried, as has intralesional interferon; superficial ablative therapies including liquid nitrogen, electrocautery, laser and LEEP have been attempted with varying success rates. Circumferential surgical resection almost inevitably results in unacceptable loss of sphincter control and soiling, but anoscopy-directed limited excision may prove less morbid. The fact that so many options are available implies perhaps that no single modality is yet considered sufficiently effective, with minimal complication [ 12 , 13 ]. This, too, is reminiscent of the early years of management of cervical pre-neoplasia, when a host of methods was pursued in attempted elimination of focal lesions of the squamocolumnar junction. As in the cervix, human ingenuity will undoubtedly prevail, and one or two forms of extirpation will emerge as both efficacious and uncomplicated. An interesting consideration is whether or not anal cancer, and thereby anal cytopathology screening programs, would be of value in the developing world, particularly in Africa, India and China, where the bulk of the global incidence of HIV resides. This will depend on two very different factors. First is the nature of transmission. Unlike the situation in the developed world, AIDS in these regions is not essentially a disease of homosexual males; thus, without anal intercourse predominating, an upward trend in the incidence of anal cancer would seem unlikely. Anal screening programs would be unnecessary or cost- in effective in these communities. The second feature dictating the institution of screening programs in developing countries relates to antiretroviral therapy. Anal intraepithelial neoplasia and cancer are not encountered early in the progression of HIV/AIDS. Rather, they are late complications of patients living long-term with AIDS, usually implying patients living long-term on HAART [ 14 ]. Unless affordable very low cost antiretroviral drugs could be manufactured and distributed widely, it seems unlikely that patients in developing countries would survive into the time zone in which delayed neoplasms such as anal cancer become a public health priority.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC552334.xml
552308	Quality of life in patients with personality disorders seen at an ordinary psychiatric outpatient clinic	Background Epidemiological studies have found reduced health-related quality of life (QoL) in patients with personality disorders (PDs), but few clinical studies have examined QoL in PDs, and none of them are from an ordinary psychiatric outpatient clinic (POC). We wanted to examine QoL in patients with PDs seen at a POC, to explore the associations of QoL with established psychiatric measures, and to evaluate QoL as an outcome measure in PD patients. Methods 72 patients with PDs at a POC filled in the MOS Short Form 36 (SF-36), and two established psychiatric self-rating measures. A national norm sample was compared on the SF-36. An independent psychiatrist diagnosed PDs and Axis-I disorders by structured interviews and rated the Global Assessment of Functioning (GAF). All measurements were repeated in the 39 PD patients that attended the 2 years follow-up examination. Results PD patients showed high co-morbidity with other PDs and Axis I mental disorders, and they scored significantly lower on all the SF-36 dimensions than age- and gender-adjusted norms. Adjustment for co-morbid Axis I disorders had some influence, however. The SF-36 mental health, vitality, and social functioning were significantly associated with the GAF and the self-rated psychiatric measures. Significant changes at follow-up were found in the psychiatric measures, but only on the mental health and role-physical of the SF-36. Conclusion Patients with PDs seen for treatment at a POC have globally poor QoL. Both physical and mental dimensions of the SF-36 are correlated with established psychiatric measures in such patients, but significant changes in these measures are only partly associated with changes in the SF-36 dimensions.	Background According to the DSM-IV [ 1 ] personality disorders (PDs) are characterized by enduringly deviating patterns of perceiving, relating to, and thinking about the environment and oneself that are exhibited in a wide range of social and personal contexts. Such patterns lead to "clinically significant distress or impairment in social, occupational, or other important areas of functioning". The DSM-IV does not indicate how "clinically significant distress or impairment" (page 633) should be evaluated, however, and a recent study showed that various formulations of this criterion hardly increased diagnostic validity [ 2 ]. Since the DSM-IV included the Global Assessment of Functioning Scale (GAF) as Axis V, it is reasonable to consider if the GAF should be used for evaluations of "significant distress or impairment". Two problems are implicit in such an approach: at what GAF cut-off score should "significant distress or impairment" be set, and the evaluation is only done by a professional. As to the first problem, Kessler et al [ 3 ] suggested a GAF score of 60 as a cut off for "serious mental illness". As to the second, in patients with somatic diseases "clinically significant distress or impairment" has for a long time been quantified by self-rating of the health-related quality of life (QoL) including mental health [ 4 ]. Several general instruments for the rating of QoL have been developed among which the Medical Outcome Study Short Form-36 (the SF-36) and its brief version the SF-12 have become the most popular [ 5 , 6 ]. Major treatment outcome studies of PDs like the Collaborative Longitudinal Personality Disorders Study [ 7 ], the Norwegian Network of Psychotherapeutic Day Hospitals [ 8 ], and the Cassel Hospital study [ 9 ] used the GAF, however, and did not included any QoL measurements, and the QoL was not included in a recommended core battery of instruments for measurement of changes in PD patients [ 10 ]. In contrast, two epidemiological studies of PDs have used the SF-12 as a measure of disability. In a national study from Australia, Jackson and Burgess [ 11 ] reported that the SF-12 Physical and Mental Component Summary Scales (PCS and MCS) were both significantly reduced in persons with one or more PDs diagnosed by a screening instrument, compared to persons without. When further examining the relationship between the SF-12 and PDs, they found that co-morbid Axis I and chronic physical conditions explained a considerable part of the MCS and PCS scores in PDs [ 12 ]. Mean MCS and PCS scores also became significantly lower with an increasing number of comorbid PDs present. A national study from the United States [ 13 ] confirmed the reduced MCS after controlling for co-morbid Axis I disorder in the avoidant, dependent, paranoid, schizoid, and antisocial PDs, but not in the histrionic PD, diagnosed by a more extensive diagnostic interview schedule than the Australian study. A review of the literature showed that the QoL had only been used as a disability measure in a few clinical research studies of PD patients. In depressed elderly patients, Abrams et al [ 14 ] found that the presence of criteria for cluster B PDs predicted lower QoL. Since the cluster B criteria overlapped considerably with symptoms of depression, it was unclear if they made any independent contribution to reduced QoL. Swinton et al [ 15 ] reported that male PD patients in a high security forensic setting were less satisfied with their overall QoL than patients with schizophrenia. The authors emphasized that the high security setting was quite unusual. Hueston et al. [ 16 ] showed that primary care patients with high risk for PDs, scored significantly lower on overall QoL and on several subscales of the SF-36, compared to patients with a low risk for PDs. Since prevalence of depression and alcohol dependence was higher in the high-risk group, the influence of PDs alone on QoL was difficult to tease out in that study. Nakao et al [ 17 ] examined the relationship between PDs and the GAF in 136 Axis I patients mainly with mood and anxiety disorders and found that patients with any comorbid PDs were more disabled than those without. They did not adjust for the presence of Axis I disorders, however. None of these clinical studies takes QoL as observed in PD patients seen at an ordinary psychiatric outpatient clinic (POC) as their point of departure. However PD patients are frequent at POC, and the QoL is an important self-rated measure of "clinically significant distress or impairment". Since QoL data on PD patients seen at at POC seems to be lacking from the literature, we found it relevant to study a consecutive sample of PD patients from a POC and collect QoL data with SF-36. We posed the following research questions: 1) How are the SF-36 dimensions mean scores in PD patients compared to age- and gender-adjusted norm data? 2) To what extent are the SF-36 scores in PD patients associated with co-morbid Axis I disorders? 3) How is the association between the SF-36 dimension and established patient- and professional-rated psychiatric measures in PD patients? and 4) What changes in the SF-36 dimensions of treated PD patients are observed from baseline to follow-up, and how are they related to changes in the psychiatric measures? Methods Setting Furuset POC serves a communality of Oslo City, Norway with a population of 28.000 people. At the time of the study, the staff consisted of three psychiatrists, three clinical psychologists, two psychiatric nurses, and two social workers. The intake rate was approximately 400 new patients a year. The first author (KN) invited the staff to take part in the study by referring to her new patients with probable PDs. Six professionals were willing to participate, while four declined due to heavy clinical burden, or lack of interest. At the start of the study in 1996, Furuset was a new suburb of Oslo, and the inhabitants were characterized by lower socio-economic conditions, high mobility, and a considerable prevalence of immigrants from Asian countries. The suburb had a high proportion of municipal housings, and the criterion for allotment to them was severe mental disorder and/or severe socio-economic problems. Many patients seen at Furuset POC were out of work due to mental disorders, and/or due to socio-economic circumstances. Patients Patients aged from 18 to 75 years were consecutively recruited from January 1, 1996 to June 30, 1998. The patients were referred from the local GPs, and physical examination and adequate treatment and follow-up of physical diseases were the responsibility of the GPs. The six therapists screened for probable PDs among new patients scheduled for treatment. Exclusion criteria were mental retardation, lifetime psychosis and bipolar disorder, organic mental disorders, current strong suicidal ideation, and insufficient knowledge of the Norwegian language. Eligible patients received oral and written information about the study from their therapists. Then the patients were invited to take part in the study, and they all gave written informed consent. The Ethical Review Board of Department of Psychiatry, Aker University Hospital approved the project. The six therapists did not miss out any patients at screening, but 5 (4%) eligible patients declined to take part in the study. Among 110 eligible patients referred to the study, only 91 filled in the SF-36 at baseline due to administrative misunderstandings. However, when they were compared to the 19 who did not fill in, the non-attenders only had significantly fewer co-morbid Axis I-disorders (data not shown). In order to answer the research questions, the sample was divided into three groups: cluster A+B PDs (n = 39), cluster C (n = 33), and Axis I-disorders (n = 19). The cluster A+B group could also contain co-morbid cluster C PDs and Axis I-disorders, and the cluster C co-morbid Axis I-disorders. Follow-up procedure Two years after baseline, the patients received a mailed written appointment for a follow-up interview. Those who did not show up were sent a written reminder. If they still did not meet, they were called by phone, and if there was no answer, their addresses and phone numbers were checked at the Census register. Appointments were mailed to new addresses, and phone-calls were made in case of non-response. Only a few patients responded to these extended search procedures. Norm sample Norm data on the SF-36 was obtained from the Survey of Level of Living in Norway 1998 [ 18 ] comprising 6.638 participants aged 23 to 75 years. The norm data were adjusted by gender and distribution into 5-year age groups in relation to the PD sample. Assessments At baseline, diagnoses of PDs were made with the use of the Personality Disorder Examination, and Axis I-disorders were diagnosed by the MINI-International Neuropsychiatric Interview. Anamnestic data were collected, and global assessment of function was rated. The professional-based interviews and examinations of all patients at baseline and follow-up were carried out by a single experienced psychiatrist (KN), who did not take part in any treatment given. All patients also filled in the following self-rating instruments at baseline: the SF-36, the Social Adjustment Scale, and the Symptom Checklist 90-Revised Personality Severity Index. At follow-up all these assessments were repeated, and additional information about treatment as well as job/education, social- and family changes was collected. Measures Professional-rated The Personality Disorder Examination (PDE) [ 19 ] is a structured clinical interview for PDs according to the DSM-III-R with good inter-rater reliability, and wide international application. Findings are reported as PD diagnoses, and as dimensional PD scores based on the sum of the scoring on each PD criterion (0: not present, 1: probably present, and 2: definitely present). Dimensional scores for the PD clusters are used as a main psychopathology variable, and the numbers of PDs are also reported. The MINI International Neuropsychiatric Interview [ 20 ] was used to diagnose Axis-I disorders according to DSM-IV. The MINI covers 18 Axis-I disorders, has been translated into many languages and has demonstrated good inter-rater reliability. Findings are reported as numbers and percentages of patients with positive Axis-I diagnoses, and as mean number of such diagnoses. The Global Assessment of Functioning (GAF) is a rating scale for the current evaluation of the overall functioning of a subject on a continuum from severe mental disorder to complete mental health that was defined as Axis V of the DSM-IV. Scale values range from 1 (sickest individual) to 100 (the healthiest person). The scale is divided in ten equal intervals from 1 – 10 to 91 – 100. Most outpatients will be rated between 40 and 70, although some individuals rated above 70 may seek therapy. The GAF is a reliable instrument [ 21 ], and the cut-off score for 'minimal impairment' has been set at 70 points or higher [ 22 ] and for 'serious mental disorder' at lower than 60 [ 3 ]. Patient-rated The SF-36 [ 5 ] was chosen for measurement of health-related QoL, since it is in widespread use, and has shown good psychometric properties in Norway [ 23 ]. The SF-36 has demonstrated sensitivity to change, and score changes can be interpreted as changes in the health-related quality of life of the patient. The SF-36 assesses eight dimensions of physical and mental health, and the range is from 100 (optimal) to zero (poorest): physical functioning (PF), physical role functioning (RP), bodily pain (BP), general health (GH), vitality (VT), social functioning (SF), emotional role functioning (RE), and mental health (MH). The Social Adjustment Scale (SAS-SR) [ 24 ] contains 42 questions which investigate expressive or instrumental roles in six major areas of functioning: work, social and leisure activities, relationship with extended family, role as a spouse/ partner, role as a parent, and role as member of the family unit. Each area is measured as a continuous variable, and the scales for individual items range from 1 (best) to 5 (worst). The scores within each role area are summed, and a mean for each area is obtained. By adding up the scores of all items and dividing by the number of items actually scored, an overall adjustment score is obtained. The Symptom Checklist 90-Revised (SCL-90-R), Personality Severity Index (PSI). The SCL-90-R is a 90-item self-report inventory assessing current levels of mental symptoms patterns. Each item is a description of a mental symptom rated on a five-point scale, and rates the degree of 'distress/discomfort' during the last week prior to its administration. Several indices based on the SCL-90-R scores have been defined, and the PSI is the mean value of 22 items covering the interpersonal sensitivity, anger/ hostility, and paranoid ideation subscales. The PSI reflects the presence and severity of relatively enduring characteristics of the patient, and is, therefore, relevant for the evaluation of severe PDs [ 25 ]. For the PSI, pathology is defined by a cut-off score of ≥ 1.0. Statistics Data were analyzed by SPSS version 12.0. Descriptive statistics were conducted with independent and paired-sample t-test as well as one-way ANOVA (with Bonferroni's correction for multiple comparisons) for metric variables, and with χ 2 and Fisher's Exact Test for categorical variables. The Mann-Whitney test and the Wilcoxon signed-rank test were applied for metric variables when the data distribution violated parametric assumptions. Spearman's rank correlation was used for associations. One-sample t-tests were applied when the SF-36 dimension scores of the age- and gender-adjusted norm groups were compared to the means of the three diagnostic groups. The comparison of mean scores on the SF-36 dimensions between the three groups was conducted with oneway ANOVA with Bonferroni's correction. The influence on the QoL scores of patients with PDs of Axis I disorders and of cluster C PDs in the cluster A+B PDs group, was examined with linear regression analyses. All tests were two-sided, and the level of significance was set at p < .05. Results Sample characteristics All the 91 patients included were Caucasian, and 48 (53%) were females. The mean age of the sample at baseline was 36.3 years (SD 10.5) and ranging from 19 to 74 years, with a median of 35 years. None of the patients had any significant somatic diseases as reported by their GPs. Patients belonging to the cluster A+B, cluster C, and Axis I- disorders groups did not differ at baseline on demographic variables (Table 1 ). Table 1 Demographic and psychopathological features at baseline in patients with cluster A+B, and cluster C personality disorders, and non-psychotic Axis I disorders. Variables Cluster A+B (n = 39) Cluster C (n = 33) Axis I (n = 19) p Age (mean, SD) 35.1 (11.2) 36.6 (10.9) 41.1 (10.4) .15 Gender (n, %) .74 Male 17 (44) 17 (52) 8 (42) Female 22 (55) 16 (48) 11 (58) Relationship (n, %) .53 Paired 23 (59) 18 (55) 7 (37) Non-paired 16 (41) 15 (45) 12 (63) Basic education level (n, %) .28 ≤ 9 years 15 (33) 13 (32) 8 (33) 10 – 12 years 14 (31) 16 (39) 13 (54) ≥ 13 years 16 (36) 12 (29) 3 (13) Early childhood loss (n, %) 12 (31) 7 (21) 1 (5) .22 Childhood sexual abuse (n, %) 7 (18) 7 (21) 1 (5) .22 Age of onset mental problems (mean, SD) 16.7 (8.9) 20.0 (8.4) 27.0 (11.8) <. 001 A+B, C vs Axis I Work income (n, %) 23 (59) 21 (64) 13 (68) .78 Economic support 16 (41) 12 (36) 6 (32) Income last year (1.000 NOK) (mean, SD) 154 (74) 178 (70) 179 (62) .27 On sickleave last year (n, %) .37 No 18 (46) 10 (30) 5 (26) ≤ 12 weeks 11 (28) 12 (37) 5 (26) ≥ 13 weeks 10 (26) 11 (33) 9 (52) No of PD cluster criteria (mean, SD) Total 41.8 (16.2) 26.6 (11.9) 8.1 (9.2) < .001 A+B vs C vs Axis I Cluster A 11.3 (8.2) 5.6 (4.6) 1.6 (2.0) < .001 A+B vs C, Axis I Cluster B 14.7 (10.9) 4.8 (5.4) 2.6 (3.7) < .001 A+B vs C, Axis I Cluster C 15.8 (10.0) 16.2 (5.7) 3.9 (5.4) < .001 A+B, C vs Axis I Comorbid Axis I disorder (n, % .60 No 10 (26) 8 (24) - Yes 29 (74) 25 (76) GAF (mean, SD) 41.5 (9.1) 51.0 (7.4) 55.4 (9.3) < .001 A+B vs C, Axis I PSI (mean, SD) 1.80 (.78) 1.24 (.71) 1.24 (.81) .005 A+B vs C, axis I SAS (mean, SD) Overall adjustment 2.70 (.66) 2.59 (.59) 2.16 (.52) .007 A+B, C vs Axis I Work 2.86 (1.62) 3.01 (1.80) 2.07 (.50) .94 Social and leisure 3.18 (1.18) 2.96 (1.06) 2.44 (.91) .06 Extended family 2.29 (.60) 2.00 (.51) 1.95 (.68) .05 Marital, partnership 2.21 (.56) 2.59 (.78) 2.14 (.63) .10 Parental 2.00 (.66) 1.91 (.58) 1.52 (.67) .14 Family unit 2.17 (.96) 2.21 (.95) 2.10 (.81) .94 As to psychopathology, the Cluster A and B PDs criteria sum scores were significantly higher in the Cluster A+B group compared to the two other groups, and both cluster A+B and cluster C group had significantly higher scores on cluster C criteria sum score than the Axis I group (Table 1 ). The Cluster A+B group had significantly lower GAF-score and higher PSI score than the two other groups which did not differ from each other. Mental problems had started significantly earlier in the PD groups compared to the Axis I-group. The SAS Overall adjustment was significantly poorer in the cluster A+B and cluster C groups, compared to the Axis I group. The diagnostic distribution of PDs and Axis I disorders in the three groups at baseline and follow-up are given in Table 2 . The mean number of PDs diagnosed in cluster A+B was 2.2, and in cluster C 1.3 per patient, and the most common PDs were avoidant, borderline, and dependent. One PD was observed in 37 patients (51%), two PDs in 20 (28%), and three PDs in 15 patients (21%). Fifty-four (75%) of the PD patient had at least one co-morbid Axis I-disorder. Table 2 Diagnostic distribution of the baseline and follow-up samples. Baseline Follow-up Cluster A+B (n = 39) Cluster C (n = 33) Axis I (n = 19) Cluster A+B (n = 18) Cluster C (n = 21) Axis I (n = 11) Personality disorders (DSM-III-R) N (%) N (%) N (%) N (%) N (%) N (%) Paranoid 14 (36) - - 7 (39) 2 (10) 1 (9) Schizotyp 6 (15) - - 2 (11) 1 (5) - Schizoid 4 (10) - - 4 (22) - - Antisocial 4 (10) - - 1 (6) - - Narcissistic 0 (0) - - 0 (0) - - Histrionic 4 (10) - - 0 (0) - - Borderline 19 (49) - - 2 (11) - - Dependent 5 (13) 7 (21) - 1 (6) 3 (14) - Avoidant 15 (39) 25 (76) - 11 (61) 15 (71) 3 (27) Obsessive-compulsive 5 (13) 8 (24) - 1 (6) 2 (10) - Passive-aggressive 8 (21) 2 (6) - 1 (6) 2 (10) - Mean no of PDs 2.2 1.3 - 1.7 1.2 .4 Axis I disorders (DSM-IV) Major depression 6 (15) 9 (27) 3 (16) 0 (0) 1 (5) 1 (9) Dysthymia 5 (13) 11 (33) 5 (26) 4 (22) 4 (19) 1 (9) Panic disorder 7 (18) 7 (21) 2 (11) 0 (0) 2 (10) 1 (9) Agoraphobia 5 (13) 7 (21) 2 (11) 2 (11) 2 (10) 2 (18) Social phobia 9 (23) 8 (24) 2 (11) 2 (11) 4 (19) 1 (9) GAD* 1 (3) 0 (0) 2 (11) 0 (0) 0 (0) 0 (0) OCD* 4 (10) 2 (6) 2 (11) 0 (0) 0 (0) 1 (9) Alcohol dependence 10 (26) 2 (6) 3 (16) 1 (6) 0 (0) 2 (18) Substance dependence 7 (18) 0 (0) 2 (6) 0 (0) 1 (5) 0 (0) Bulimia nervosa 4 (10) 1 (3) 0 (0) 0 (0) 1 (5) 0 (0) PTSD* 0 (0) 1 (3) 0 (0) 0 (0) 0 (0) 0 (0) Mean no of Axis I disorders 1.5 1.5 1.2 .5 0.7 0.8 GAD: Generalized anxiety disorders, OCD: Obsessive-compulsive disorder, PTSD: Post-traumatic stress disorder Axis-I disorders were equally common in the two PD groups (mean 1.5 disorder) and with slightly lower mean (1.2) in the Axis I-disorder group. Depressions, anxiety disorders, and alcohol dependence were the most frequent Axis I diagnoses in all groups. QoL in PD patients Figure 1 . shows that the mean scores on the eight dimensions of the SF-36 of the PD patients at baseline are significantly lower (p < .001 for all) than those of the age- and gender-adjusted norms. The mean difference was least (13 points) for PF, and highest for RF and RE (54 points and 49 points, respectively). Figure 1 SF-36 mean dimensional scores in personality disorder sample (N = 72) and the age- and gender-adjusted norm sample. Linear regression analyses showed that control for co-morbid Axis I disorders reduced the PF, GH, VT, SF, and MH scores of the total PD group significantly. Controlling for comorbid cluster C PDs did not influence the SF-36 scores of the cluster A+B PDs to any significant extent, while control for Axis disorders significantly reduced the MH scores in the cluster A+B and cluster C groups. No significant differences were found between genders on any of the SF-36 dimensions among the PD patients (data not shown). Both the cluster A+B, the cluster C, and the Axis I group differed significantly from their norms on all eight SF-36 dimensions (data not shown). No significant differences were observed on the eight SF-36 dimensions between the three diagnostic groups (Figure 2 ). Figure 2 SF-36 mean dimensional scores at baseline for Cluster A+B, Cluster C, and Axis I-disorders groups. When we compared the patients with one (n = 37), two (n = 20), and three or more (n = 15) PDs, we did not observe any significant differences in mean MCS and PCS scores. Correlation between SF-36 dimensions and other measures The eight dimensions of SF-36 are regularly divided into the four physical: PF, RP, BP, and GH, and the four mental ones: VT, SF, RE, and MH. In our PD sample the SF-36 mental dimensions had most significant correlations with the psychiatric measures of the GAF, the SCL-90-R PSI, the sum of positive PDs diagnostic criteria, and the dimensions of the SAS (Table 3 ). The SF-36 MH had a significant correlation to most of these measures, followed by VT and SF. The physical dimensions of the SF-36 had less frequently a significant correlation to the psychiatric measures. Table 3 Correlation of SF-36 dimensions with Global Assessment of Functioning (GAF), Personality Severity Index (PSI), and dimensions of Social Adjustment Scale (SAS). PF RP BP GH VT SF RE MH GAF .12 .32* .02 .32** .26* .36 .30 .40 Total no of positive PD criteria .03 -.11 .02 -.27 -.14 -.18 -.05 -.26** SCL-90-R PSI -.22* -.17 -.12 -.26* -.25* -.34 -.19 -.38 SAS Overall -.38 -.31 -.23* -.34 -.42 -.45 -.30 -.45** SAS Work -.17 -.22* -.10 -.16 -.44** -.31* -.26* -.41 SAS Social and leisure -.30 -.35** -.24* -.32 -.43 -.39 -.29 -.37** SAS Extended family -.20 -.18 -.23* -.21 -.26* -.32* -.28* -.43 SAS Marital/partner -.18 .01 -.04 -.18 -.24 -.08 -.05 -.10 SAS Parental -.04 -.06 .06 .02 -.07 -.09 -.07 -.04 SAS Family unit -.38 -.22 -.26* -.20 -.37** -.31* -.27* -.35** Sum significant correlation 4 4 4 4 6 7 6 8 * Correlation is significant at the .05 level (two-tailed), ** Correlation is significant at the .01 level (two-tailed) The SAS overall adjustment and the SAS social and leisure functioning had significant correlations to all the SF-36 dimensions, while the SAS marital/ partnership and the SAS parental functioning had none. The SAS work, extended family, and family unit fell in between. What changes in the QoL of PD patients can be observed from baseline to follow-up two years later? Although quite intensive search for patients was done for the follow-up examination, only 50 patients (53%) of the 91 patients who rated themselves on the SF-36 complied. The distribution of patients were cluster A+B group (n = 18), cluster C (n = 21), and Axis I-disorder group (n = 11). Due to small sample sizes, the Axis I disorder was dropped from further analysis and the two PD groups were pooled as to the study of changes after treatment. The 39 PD patient with SF-36 ratings both at baseline and follow-up were compared to the 33 PD patients only seen at baseline. The diagnoses at follow-up are shown in Table 2 , and among the non-compliant patients those with borderline PD, and alcohol and substance dependence were over-represented. Few significant differences were observed between compliant and non-compliant PD patients at follow-up (Table 4 ). In particular, no significant differences of the eight SF-36 dimensions were observed between the compliers and non-compliers at baseline. The compliers had significantly more depressive disorders and cluster C PDs at baseline. All of those who terminated treatment without the consent of their therapist (N = 22) were in the non-compliant group. The non-compliant patients also had a significantly longer mean duration of treatment. The mean treatment time for the PD patients attending follow-up was 16.6 months (SD 5.9), median 18 months, and range 4 to 24 months, and the mean follow-up time since treatment termination was 9.8 months (SD 6.4), median 10.4 months, and range 0 to 26 months. The majority of the patients had weekly individual psychotherapy, although a small proportion also had group psychotherapy in addition. Drug treatment was given to 20 patients of the 39 patients, and to 20 of the 33 non-compliers (ns). Among those seen at follow-up, had 15 got antidepressive and 5 antipsychotic medication, in addition to psychotherapy. Table 4 Demographic, psychopathological, and treatment features at baseline for patients with personality disorders with (N = 39) and without (n = 33) follow-up examination. Variable Follow-up + (n = 39) Follow-up - (n = 33) p Age (mean, SD) 37.9 (11.6) 33.3 (9.9) .08 Gender (n, %) .50 Male 17 (44) 17 (52) Female 22 (56) 16 (48) Relationship (n, %) .56 Paired 18 (46) 13 (39) Non-paired 21 (54) 20 (61) Basic education level (n, %) .64 ≤ 9 years 11 (29) 13 (40) 10 – 12 years 13 (34) 10 (30) ≥ 13 years 14 (37) 10 (30) ≥ 1 cluster A PDs 13 (33) 6 (18) .15 ≥ 1 cluster B PDs 10 (26) 15 (45) .08 ≥ 1 cluster C PDs 35 (90) 22 (67) .02 Mean (SD) of PD cluster criteria Total 34.8 (17.9) 34.8 (14.3) .99 Cluster A 9.5 (8.7) 7.7 (5.1) .28 Cluster B 7.6 (9.3) 13.1 (10.3) .02 Cluster C 17.6 (8.9) 14.0 (7.2) .07 ≥ 1 depressive disorder 19 (49) 7 (21) .02 ≥ 1 anxiety disorder 17 (44) 16 (49) .68 ≥ 1 substance use disorder 9 (23) 11 (33) .33 Comorbid Axis I disorder (n, %) .79 No 9 (23) 9 (27) Yes 30 (77) 24 (33) GAF (mean, SD) 46.0 (9.4) 45.7 (9.9) .91 PSI (mean, SD) 1.5 (.9) 1.6 (.7) .85 SAS Overall (mean, SD) 2.7 (.6) 2.6 (.6) .86 SF-36 (mean, SD)* Physical Functioning 79.4 (19.1) 76.7 (22.5) .72 Role Functioning 31.4 (34.3) 25.8 (36.2) .27 Bodily Pain 47.6 (28.9) 48.1 (26.7) .71 General Health 51.4 (23.5) 50.5 (20.39 .99 Vitality 35.0 (19.6) 29.1 (18.6) .16 Social Functioning 45.2 (28.5) 48.9 (21.5) .60 Role Emotional 42.7 (39.7) 32.3 (31.7) .35 Mental Health 42.5 (23.2) 38.3 (19.5) .52 No of sessions (mean, SD)* 16.6 (5.9) 18.8 (26.9) .01 Termination without consensus (n, %) 0 (0.0) 22 (67) < . 001 Treated by specialist (n, %) 14 (36) 16 (49) .28 Additional drug treatment (n, %) 20 (51) 20 (61) .43 *Mann-Whitney tests In the 39 PD patients who complied at both baseline and follow-up, significant improvement was seen in the RF and MH dimensions of the SF-36, while considerable, but non-significant changes were observed for BP and SF (Table 5 ). Table 5 Changes from baseline to follow-up in patients with personality disorders (n = 39). Measure Baseline Mean (SD) Follow-up Mean (SD) P SF-36 Physical Functioning 79.4 (19.2) 76.8 (24.6) .95 Role Physical 31.4 (34.3) 51.3 (38.5) .01 Bodily Pain 47.6 (28.9) 57.5 (25.6) .06 General Health 51.4 (23.5) 56.0 (26.6) .22 Vitality 35.0 (19.6) 36.3 (21.4) .70 Social functioning 45.2 (28.5) 53.5 (28.7) .09 Role-emotional 42.7 (39.7) 41.9 (38.0) .89 Mental Health 42.5 (23.2) 50.1 (22.3) .03 Global Assessment of functioning 46.0 (9.4) 54.6 (9.6) < .001 Total no of PD criteria 34.8 (17.9) 25.7 (11.5) < .001 SCL-90-R PSI 1.52 (.86) 1.30 (.80) .035 Social Adjustment Scale (SAS) Overall adjustment 2.66 (.63) 2.42 (.62) .007 Work 2.76 (1.49) 2.36 (1.39) .20 Social and leisure 3.17 (1.22) 2.87 (1.11) .045 Extended family 2.05 (.52) 1.97 (.50) .34 Both professional-rated measures the GAF, and the mean total number of PD criteria, showed significant improvement. Among the patient-rated measures significant better results at follow-up were found for the SCL-90-R PSI, the SAS overall adjustment, and the SAS social and leisure scales. Discussion The main findings of this study of mainly co-morbid PD patients treated at an ordinary POC, was that the QoL on both the physical and mental SF-36 dimensions was significantly lower than that of an age- and gender-adjusted general population sample. According to our knowledge, ours is the first report on QoL-data in such PD patients at a POC. This finding is in accordance with QoL studies of PD patients in the general population [ 11 , 13 ], and correspond to findings of clinical studies of patients with anxiety disorders, depression, schizophrenia, and substance dependence [ 27 - 30 ]. However, the SF-36 dimension mean scores of our PD sample are lower than those reported for these diagnoses, and for co-morbid disorders [ 31 ]. In our sample we did not find any significant differences between the SF-36 dimension scores of the cluster A+B, cluster C, or Axis I groups, and all groups had significantly lower scores on all dimensions than their age- and gender-adjusted norm groups. In contrast to the epidemiological study from Australia [ 12 ] we did not find worsening of MCS and PCS with increasing number of PDs present in our sample. This could be due to our small samples, but also due to the fact that our patients with 1 PD had considerably lower QoL than in the Australian survey [MCS: 33.7 (SD 10.6) versus 44.4 (SD 12.0), p < .001, and PCS: 43.8 (SD 8.6) versus 46.9 (SD 11.0), p = .03]. Comorbid Axis I disorders explained a significant part of scores of PF, GH, VT, SF, and MH scores of the total PD group. This is in accordance with the findings of the Australian study [ 12 ]. We found that the SF-36 dimensions had variable associations with established psychiatric measures. As expected the SF-36 MH was most strongly associated with the psychiatric measures, but so were also SF and VT. For the SAS we found that overall adjustment and social and leisure activities were significantly correlated to all the SF-36 dimensions. In our PD sample we observed a somewhat different pattern of significant correlations between the GAF and the SF-36 dimensions than reported by Meijer et al. [ 32 ] in patients with schizophrenia. Small sample sizes and different diagnostic classes could be the explanation. However, in sum the SF-36 had a considerable association with established psychiatric measures in our PD sample. For both the patient- and professional-rated psychiatric measures significant changes at follow-up after treatment was observed in the 39 patients who also scored themselves on the SF-36. We cannot say if these changes were related to treatment, and ours is not an outcome study. We wanted to examine if changes in established psychiatric measures were associated with changes in the QoL measured by the SF-36 in the PD patients seen at a POC. Significant changes at follow-up were found for only two of the SF-36 dimensions, however, one physical (RP) and one mental (MH). While the finding for MH was expected, the change in RP which covered problems with work or other daily activities as a result of physical health was more difficult to explain. The score on that dimension was extraordinarily low at baseline (mean 31.4), and regression towards mean could be a likely explanation. It seemed that only MH of the SF-36 changed in the same way as established psychiatric measures in our study. The SF-36 MH correlated significantly with most of such psychiatric measures, and MH is currently used as a valid measure for mental health in several studies [33]. This result could indicate that the other dimensions of the SF-36 are less valid as measures of changes in mental health of PD patients, or alternatively that most aspects of QoL measured by the SF-36 do not change in PD patients even if established psychiatric measures do. The main strength of our study was that we were able examine systematically various aspects of the QoL measured by the SF-36 in a clinically relevant sample of PD patients at a POC which is a common setting for such patients in psychiatry. Our study had a number of weaknesses. The study groups were small with limited statistical power, and there was a considerable risk of type II errors. More significant differences as to the SF-36 dimensions could turn up in larger samples. Although we put considerable efforts into location of patients, we had a lower follow-up rate than we had expected. However, the PD patients who did not show up at follow-up did not differ much from those who did. We cannot, therefore, generalize the discrepancy observed between significant changes in established psychiatric measures and lack of such changes in most of the SF-36 dimensions of PD patients treated at a POC to widely. The same experienced psychiatrists did all the interviews at baseline and follow-up. Although she was not involved in any treatments, we cannot exclude an expectation bias from her side. We think that our study has to be considered an exploratory one. Our finding of a generally strongly reduced QoL should be replicated in a PD sample with less comorbid Axis I disorders, although their influence was limited. The same is true for QoL as a valid measure for change in PD patient, since it was not recommended as part of a standard outcome battery and was not used by major treatment studies of PD patients. However, our study confirmed that the SF-36 MH dimension seemed to be a valid psychiatric measure in our PD patient sample. Conclusion In this study of the QoL in PD patients seen at an ordinary POC, we found that the PD patients had significantly lower mean scores on all the SF-36 dimensions compared to age- and gender-adjusted norm data. This is in accordance with the SF-36 measurements of other major diagnostic groups of mental disorders. Although the SF-36 dimensions correlated considerably with established psychiatric measures in our PD patients, they did not show the same significant changes over time as the established measures. The use of QoL measures like the SF-36 as an outcome measure in PD patients is in need of further investigation. List of abbreviations BP: SF-36 Bodily pain GAF: The Global Assessment of Functioning GH: SF-36 General health MH: SF-36 Mental health PD: Personality disorder PDs: Personality disorders PF: SF-36 Physical functioning POC: Psychiatric outpatient clinic PSI: Personality severity index of SCL-90-R QOL: Health-related quality of life RE: SF-36 Emotional role functioning RP: SF-36 Physical role SAS: The Social Adjustment Scale SCL-90-R: The Symptom Checklist 90-Revised SF-36: MOS Short Form 36 SF: SF-36 Social functioning VT: SF-36 Vitality Competing interests The author(s) declare that the have no competing interests. Authors' contributions KN conceived and planned the study, prepared the therapists at Furuset Outpatient Department, did all the psychiatric interviews at baseline and follow-up, and drafted the manuscript. AM helped designing of the study, supervised the statistic calculations, and drafted the manuscript. AAD participated in the design and coordination of the study, and drafted the manuscript. All authors read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC552308.xml
553994	Mutational study of sapovirus expression in insect cells	Human sapovirus (SaV), an agent of human gastroenteritis, cannot be grown in cell culture, but expression of the recombinant capsid protein (rVP1) in a baculovirus expression system results in the formation of virus-like particles (VLPs). In this study we compared the time-course expression of two different SaV rVP1 constructs. One construct had the native sequence (Wt construct), whereas the other had two nucleotide point mutations in which one mutation caused an amino acid substitution and one was silent (MEG-1076 construct). While both constructs formed VLPs morphologically similar to native SaV, Northern blot analysis indicated that the MEG-1076 rVP1 mRNA had increased steady-state levels. Furthermore, Western blot analysis and an antigen enzyme-linked immunosorbent assay showed that the MEG-1076 construct had increased expression levels of rVP1 and yields of VLPs. Interestingly, the position of the mutated residue was strictly conserved residue among other human SaV strains, suggesting an important role for rVP1 expression.	Introduction The family Caliciviridae is made up of four genera, Sapovirus , Norovirus , Lagovirus , and Vesivirus , which contain sapovirus (SaV), norovirus (NoV), rabbit hemorrhagic disease virus, and feline calicivirus strains, respectively. Human SaV and NoV strains are agents of gastroenteritis. The prototype strain of human SaV, the Sapporo virus, was originally discovered from an outbreak of gastroenteritis in an orphanage in Sapporo, Japan, in 1977 [ 1 ]. Chiba et al. identified viruses with the typical animal calicivirus morphology, called the "Star of David" structure, by electron microscopy (EM). SaV strains were recently divided into five genogroups (GI to GV), of which GI, GII, GIV, and GV strains infect humans, while GIII strains infect porcine species [ 2 ]. The SaV GI, GIV, and GV genomes are each predicted to contain three main open reading frames (ORFs), whereas SaV GII and GIII have two ORFs. SaV ORF1 encodes for non-structural proteins and the major capsid protein (VP1). SaV ORF2 (VP2) and ORF3 (VP3) encoded proteins of yet unknown functions. The NoV genome is organized in a slightly different way than the SaV, since ORF1 encodes all the nonstructural proteins, ORF2 encodes the capsid protein (VP1), and ORF3 encodes a small protein (VP2). Human SaV and NoV strains are noncultivable, but expression of the recombinant VP1 (rVP1) in a baculovirus expression system results in the self-assembly of virus-like particles (VLPs) that are morphologically similar to native SaV [ 3 , 4 ] In a recent NoV expression study, a single amino acid substitution in the rVP1 gene affected VLP formation but not rVP1 expression [ 5 ]. In a different study, inclusions of NoV ORF3 and poly(A) sequences in a construct increased the expression levels of NoV rVP1 and the stability of VLPs when compared to constructs without these sequences [ 6 ]. Recently, cryo-EM analysis of SaV VLPs and X-ray crystallography analysis of NoV VLPs predicted the SaV shell (S) and protruding domains (subdomains P1 and P2) that were based the NoV domains [ 7 , 8 ]. Chen et al. also described strictly and moderately conserved amino acid residues in the capsid protein among the four genera in family Caliciviridae . The purpose of this study was to compare the time-course expression of two different SaV rVP1 constructs in a baculovirus expression system by Northern blotting, Western blotting, enzyme-linked immunosorbent assay (ELISA), and EM. Our novel results have indicated that nucleotide point mutations increased the yields of SaV VLPs in insect cells, offering an alternative explanation for the increased expression levels of rVP1 and yield of VLPs. Results Wt, MQG-1076, and MEG-1076 constructs Expression of SaV rVP1 in a baculovirus expression system results in the self-assembly of VLPs [ 4 ]. However, during PCR amplification nucleotide point mutations occurred in our initial MQG-1076 construct, at nucleotide positions 4 and 1076 in VP1, which resulted in two amino acid substitutions at residues 2 and 358, respectively, and a silent nucleotide mutation at position 1895 in VP2 (Fig. 1 ). Despite these two substitutions the MQG-1076 construct formed VLPs morphological similar to native SaV (data not shown). In order to further investigate these substitutions we expressed another construct (MEG-1076 construct) having only one substitution, at residue 358 in VP1 (Fig. 1 ). This construct also formed VLPs. Finally we expressed a construct (Wt construct) without these nucleotide point mutations, i.e., having the native sequence. The Wt construct also formed VLPs, however the expression level of rVP1 was noticeably lower than those of the MQG-1076 and MEG-1076 constructs in which had similar levels (data not shown). In order to compare expression levels, we infected Wt and MEG-1076 recombinant baculoviruses each at a multiplicity of infection (MOI) of 14.5 in 2.7 × 10 6 confluent Tn5 cells in 1.5 ml of Ex-Cell 405 medium followed by incubation at 26°C. RNA transcription and rVP1 expression experiments were run in parallel for the Wt and MEG-1076 constructs. Figure 1 Schematics of the SaV constructs, Wt, MEG-1076, and MQG-1076, containing the rVP1, rVP2, and poly(A) sequences. Each construct began at the predicted AUG start. The triangles show the positions of the nucleotide point mutations. The black triangle had an amino acid substitution in the VP1, whereas the open triangle in the VP2 gene did not change amino acid sequence. An RNA probe (anti-VP1) was used to monitor the transcription of rVP1 mRNA in which contained the native sequence, i.e., lacking the mutation at 1076. Northern blot analysis Total RNA was extracted from the cells at 1, 2, 3, 4, 5, 6, 7, and 8 days postinfection (dpi) for Wt and MEG-1076 constructs. Equal amounts (500 ng) of total RNA were added to a 2% agarose gel containing formaldehyde and stained with SYBR Gold (Fig. 2A ). The rVP1 mRNA was then analysed by Northern blot with a probe specific for the VP1 gene (native sequence) corresponding to the VP1 position 157 to 1283 (Fig. 1 ). The rVP1 mRNA transcript was predicted to be approximately 2300 nucleotides long. As shown in Figure 2B , rVP1 mRNA was detected for each construct. This result showed that the insert sequence and some part of the baculovirus vector, approximately 300 nt, was transcribed, although the exact location(s) on the vector has yet to be determined. Nevertheless, the MEG-1076 construct had increased band intensities, indicating an increased steady-state level, when compared to those of the Wt construct (Fig. 2B ). For the Wt construct, rVP1 mRNA was detected at 1 dpi, peaked at 2 dpi, decreased at 3 and 4 dpi, and then decreased to undetectable levels at 5, 6, 7, and 8 dpi. For the MEG-1076 construct, rVP1 mRNA was detected at 1 dpi, peaked at 2 dpi, had steady-state levels at 3 and 4 dpi, and then decreased at 5 dpi but could still be detected at 6, 7, and 8 dpi. These results indicated that the MEG-1076 rVP1 mRNA also had greater stability when compared to those of the Wt rVP1 mRNA. Figure 2 Northern Blot analysis of Wt and MEG-1076 rVP1 mRNA. The total RNA was purified from the cells at 1, 2, 3, 4, 5, 6, 7, and 8 dpi. (A) The relative amounts of total RNA for each construct. (B) The steady-state levels of rVP1 mRNA with an anti-VP1 probe specific for the VP1 gene, corresponding to the VP1 nucleotide position 157 to 1283. Western Blot analysis Western blot analysis was used to compare the expression levels of Wt and MEG-1076 rVP1. The culture medium was separated from the cell lysate 1, 2, 3, 4, 5, 6, 7, and 8 dpi as described in the Materials and Methods. Equal volumes of culture medium and cell lysate at each dpi were used for both constructs. Proteins were separated by SDS-PAGE, electrotransferred to PVDF, and detected with a 1:3000 dilution of hyperimmune rabbit Mc114 VLP antiserum. A band at the predicted rVP1 size (60 K) was first detected in the culture medium at 2 and 4 dpi for MEG-1076 and Wt constructs, respectively, which increased each day thereafter as evidenced by an increase in band intensity (Fig. 3A ). As indicated by increased band intensities, the MEG-1076 construct expressed increased levels of rVP1 (60 K) than those of the Wt construct. Similarly, these results were reproduced using different MOIs in order to address the variability in virus stock quality (data not shown). Figure 3 Western blot analysis of Wt and MEG-1076 rVP1. Confluent Tn5 cells were infected with Mc114 recombinant baculoviruses at MOI of 14.5 and incubated at 26°C. The culture medium, including the cells, were harvested 1, 2, 3, 4, 5, 6, 7, and 8 dpi as described in the materials and methods. (A) The cell culture medium was concentrated by ultracentrifugation, resuspended in 20 μl of Grace's medium, and 5 μl was mixed with loading dye and loaded into each well. (B) The cell lysate was separated from the culture medium, resuspended in 200 μl of Grace's medium, and 5 μl was mixed with loading dye and loaded into each well. A thin band of approximately 55 K was also detected in the culture medium that appeared at 4 and 5 dpi for Wt and MEG-1076 constructs, respectively, and increased each day thereafter. In a different experiment, we determined the amino acid sequence of the MQG-1076 upper and lower bands by an Edman's degradation method. We discovered that the first three amino acid residues were MQG for both the upper and lower bands. This result indicated that the 55 K bands for these constructs were likely truncated or C-terminal deleted forms of rVP1. A thin band of 60 K was detected at every dpi in the cell lysate for the MEG-1076 construct (Fig. 3B ), however the intensity of this band did not increase to the same extent as the MEG-1076 60 K band in the culture medium (Fig. 3A ). This suggested that immediately after translation the majority of rVP1 was rapidly exported from the cells to the culture medium, though a fraction accumulated within the cells. This may also explain why no 60 K bands were detected in the cell lysate for Wt construct. The VP2 amino acid sequence was the same in all constructs. We did not detect rVP2 during the time-course expression of the MQG-1076 construct using the antiserum raised against E. coli expressed VP2 (data not shown). Antigen ELISA and EM analysis of Wt and MEG-1076 VLPs An antigen ELISA system was used to compare the yields of Wt and MEG-1076 VLPs at 1, 2, 3, 4, 5, 6, 7, and 8 dpi. The ELISA incorporated hyperimmune rabbit (capture) and guinea pig (detector) antisera raised against purified Mc114 VLPs [ 4 ]. The ELISA first detected VLPs at 2 and 3 dpi for MEG-1076 and Wt constructs, respectively (Fig. 4 ). For both constructs, the yields of VLPs increased each day thereafter, however the MEG-1076 construct had increased yields of VLPs than those of the Wt construct at 4, 5, 6, 7, and 8 dpi, approximately 6-fold increase. EM was used to verify the VLP formation of each of these constructs. We first detected VLPs at 4 dpi in the culture medium for both constructs and the numbers of VLPs increased each day thereafter (data not shown). Figure 4 Antigen ELISA analysis of Wt and MEG-1076 VLPs. The ELISA used hyperimmune rabbit (capture) and guinea pig (detector) antiserum raised against Mc114 VLPs. For the antigen ELISA, purified Mc114 VLPs were used as the positive control at concentrations ranging from 500 ng to 0.24 ng. Amino acid analysis The MEG-1076 construct contained a nucleotide point mutation in which resulted in an amino acid substitution at position 358 in VP1. We aligned 21 different VP1 amino acid sequences of SaV GI, GII, and GV strains and found this residue was strictly conserved, but more importantly, there was a strictly conserved amino acid motif at this site, NGDV (data not shown). However, when we included a porcine SaV GIII strain and a recently identified SaV GIV strain (PEC and Hou-7, respectively), only the GD site was strictly conserved, though several other amino acids nearby were also strictly conserved (Fig. 5 ). Further analysis of other SaV GIV strains are clearly needed in order to examine the possibility that the NGDV motif was moderately conserved in other human SaV strains. Figure 5 also showed that the predicted SaV P2 domain had very few conserved amino acid residues. Apart from the strictly conserved GD motif, the only other strictly conserved motif in the P2 domain was at the 5' end. Figure 5 VP1 amino acid alignment of SaV GI, GII, GIII, GV, and GV strains. We originally aligned 21 SaV GI, GII, and GV sequences but to simplify the figure we used one representative strain from each genogroup. The green bar shows the SaV P2 domain predicted by Chen et al. [7]. The asterisks indicate conserved amino acids. We originally aligned 21 different VP1 amino acid sequences of SaV GI, GII, and GV strains and found the residue (N) at position 358 (yellow) was strictly conserved (data not shown), but SaV GIII and GIV strains (PEC and Hou-7, respectively) had other residues at this position. The alignment of the five SaV genogroups showed the amino acid motif, GD, was strictly conserved (red) and several other amino acids surrounding the residue at position 358 were also strictly conserved (red). Discussion Expression of the human SaV rVP1 in a baculovirus expression system was first reported in 1997 [ 9 ]. In that study, the full-length VP1 gene, ORF2, and poly(A) sequences were included in a construct (Sapporo strain, GI). The second human SaV reported to form VLPs was with a construct (Houston/90 strain, GI) using only the VP1 sequence, i.e., lacking ORF2 and poly(A) sequences [ 10 ], while the third human SaV reported to form VLPs used a construct (Parkville strain, GI) with only VP1 and ORF2 sequences, i.e., lacking poly(A) sequence [ 7 ]. We recently expressed human SaV GI, GII, and GV rVP1 with constructs (Mc14, C12, and NK24 strains, respectively) that included ORF2 and poly(A) sequences [ 4 ]. Additional information on human SaV rVP1 expression is lacking, although it appeared that the yields of human SaV VLPs were typically low for these three genogroups. In this study, we compared the time-course expression of two different Mc114 SaV rVP1 constructs in a baculovirus expression system (Fig. 1 ). The MEG-1076 construct had two nucleotide point mutations, one in the VP1 gene in which resulted in an amino acid substitution, and one in the VP2 gene in which was silent. Although both constructs formed VLPs morphological similar to native SaV, the levels of transcription, translation, and VLP formation were clearly different. As shown in Figure 2B , the MEG-1076 rVP1 mRNA had increased steady-state levels and greater stability when compared to those of the Wt rVP1 mRNA. This difference was understood to be due to the nucleotide mutations in the MEG-1076 construct, since a similar result was observed in a NoV expression study [ 6 ]. Bertolotti-Ciarlet et al. found that a nucleotide point mutation in a NoV rVP1 construct (ORF2- A U G → A C G -ORF3+3' UTR construct, represented in bold) had decreased levels of rVP1 mRNA at 36 hours post-infection, by approximately 50%, when compared to a construct without the mutation (ORF2+ORF3+3' UTR construct). Bertolotti-Ciarlet suggested that the RNA secondary structure or changes in the mRNA stability could be responsible for the different steady-state levels, but this was not proven. Also, the MEG-1076 construct had increased levels of rVP1 expression and yields of VLPs in the culture medium when compared to those of the Wt construct (Fig. 3A ). On the other hand, the concentration of rVP1 in the cell lysate remained more or less the same during the time-course expression for the MEG-1076 construct. And for the Wt construct, rVP1 was not detected in the cell lysate, although this may have been related to the low expression levels (Fig. 3B ). Our results showed that the MEG-1076 construct had a 6-fold increase in yields of VLPs in the culture medium (Fig. 4 ), which corresponded to approximately 80 μg of CsCl purified VLPs from 200 ml of culture medium (at 6 dpi), but less than 5 μg of CsCl purified VLPs in the cell lysate (data not shown). These results suggested that either (i) immediately after translation the majority of rVP1 was exported from the cells to the culture medium where the majority of VLPs were folded but a fraction were simultaneously folded within the cells or (ii) VLPs were folded within the cells and then the majority of VLPs were immediately exported from the cells to the culture medium, though a fraction remained within the cells. In a recent NoV expression study, a single amino acid substitution in the rVP1 gene affected VLP formation but not rVP1 expression [ 5 ]. In that study, a (native) histidine residue at position 91 (relative to NoV Snow Mountain Virus strain amino acid VP1 sequence) was found to be essential for VLP formation and a construct with a substituted (mutant) arginine residue at this position failed to form VLPs despite expressing rVP1. Interestingly, that study found a single amino substitution was critical for the formation of VLPs, whereas our results showed that a single amino acid substitution was beneficial, i.e., increased the yields of VLPs. Bertolotti-Ciarlet found that inclusions of NoV ORF3 and poly(A) sequences in a construct increased the expression levels of NoV rVP1 and the stability of VLPs when compared to constructs without these sequences; and suggested that expression of other caliciviruses (NoV and SaV) rVP1 that resulted in low yields or unstable VLPs may be due to constructs that lacked the VP2 gene [ 6 ]. An alternative explanation was that point mutations influenced steady-state levels of mRNA and stability, which in turn influenced VLP formation. In our case, one or two nucleotide point mutations caused an enhancement of transcription, leading to increased yields of SaV VLPs in insect cells. Furthermore, many of these studies that expressed calicivirus rVP1 in insect cells only examined rVP1 expression and yields of VLPs but not rVP1 mRNA transcription [ 11 - 14 ]. However, another reason for the increased yields of VLPs may be associated with adaptation of SaV rVP1 to the baculovirus expression system and insect cells, since a similar result was observed with porcine enteric calicivirus in primary kidney cells [ 15 ]. Although the growth rate and replication efficiency of the recombinant baculoviruses themselves and differences in the levels of virus replication might account for such variation, we observed similar results using other MOIs, that is, the MEG-1076 construct continued to express greater yields of VLPs than the Wt construct (data not shown). Another explanation may have been differences in the extents to which these baculoviruses induce apoptosis and all these may result from features in the baculovirus skeleton rather than from the inserted SaV sequence. Such effects might for instance affect the number of adherent cells harvested or the degradation rates of both proteins and RNAs. However, we found that the MQG-1076 construct, developed from a separate experiment, had similar expression levels to that of the MEG-1076 construct (data not shown), which may eliminate the possibility that the baculovirus skeleton played a role in the increased yields of VLPs. On the other hand, we could not demonstrate whether the nucleotide mutations in VP1 and/or in ORF2 affected the transcription, a construct with only one of these mutations would be needed. Nevertheless, our results indicate that translation was exclusively affected by the single amino acid substitution in VP1. Therefore, the final increase in yields of VLPs may have been coupled at multiple levels, involving one or both of the nucleotide mutations in VP1 and VP2. We did not detect rVP2 during the time-course expression of the MQG-1076 construct (data not shown). The Wt and MEG-1076 constructs had an identical amino acid sequence, which would suggest a similar negative-result. NoV studies have found that inclusion of VP2 increases the stability of VLPs, though the expression level of NoV rVP2 was low [ 6 ]. These results may suggest that (i) SaV rVP2 was expressed at undetectable levels, (ii) SaV rVP2 was not expressed in the insect cells, or (iii) SaV rVP2 was degraded in the insect cells. The SaV GI, GIV, and GV genomes are each predicted to encode a third ORF (ORF3) overlapping the VP1 gene, whereas SaV GII and GIII have only two ORFs. The functions of SaV ORF2 and ORF3 still remain unknown. The amino acid substitution (N → S) for the MEG-1076 construct occurred in the VP1 gene at residue 358. This asparagine residue was recently identified as a moderately conserved residue among the caliciviruses capsid proteins [ 7 ], but more importantly, the residue was strictly conserved among 21 different SaV GI, GII, and GV strains and belonged to a strictly conserved amino acid motif, NGDV (Fig. 5 ). However, when we included SaV GIII and GIV strains (PEC and Hou-7, respectively) we found that only the GD amino acids were strictly conserved though several other amino acids nearby were also strictly conserved (Fig. 5 ). These data further suggested that this site played an important role in the regulation of SaV VLP formation. Recently, the cryo-EM analysis of SaV was determined and compared to NoV X-ray crystallography structure [ 7 ]. Chen et al. analysed 30 different VP1 amino acid sequences of calicivirus strains belonging to the four genera in the family Caliciviridae and identified strictly and moderately conserved residues, and predicted the P1 and P2 domains of SaV VP1 based on NoV X-ray crystallography structure. Based on these predictions, the residue at position 358 (amino acid sequence) was found as a moderately conserved residue among the caliciviruses. This arginine residue was predicated to be in the P2 domain, which is defined as the outer most protruding domain for NoV and thought to provide strain diversity [ 16 ]. Further high-resolution structural analysis of SaV VLPs is clearly needed in order to determine the precise domains and regions of SaV. However, our expression results have indicated that only approximately 80 μg of purified VLPs from 200 ml of culture medium was possible (data not shown), thus in order to determine the X-ray crystallography structure of SaV, a minimum increase in expression level of about 20-fold would be required: a challenging feat. Materials and methods Virus strain, RNA extraction, cDNA synthesis SaV GI Mc114 strain (GenBank accession number, AY237422) was isolated from a male infant seven months of age from the McCormic Hospital, Chiang Mai, Thailand on the 7th May 2001 [ 17 ]. RNA extraction and cDNA synthesis were performed as previously described [ 18 ]. PCR and sequencing Our initial SaV rVP1 construct (MQG-1076 construct) was amplified with ExTaq DNA polymerase. However, this construct was later found to have two nucleotide point mutations in ORF1 at positions 4 ( G AG → C AG) and 1076 (A A T → A G T) and one nucleotide point mutation in ORF2 at position 1895 (GT G → GT A ) (relative to the VP1 start and represented in bold). Primer and PCR errors likely introduced these mutations. These three nucleotide point mutations resulted in two amino acid substitutions in the VP1 gene, one at the second residue, where glutamic acid (E) → glutamine (Q), and one at residue 358, where asparagine (N) → serine (S). The nucleotide point mutation in ORF2 did not result in an amino substitution. Despite the two amino acid substitutions, the MQG-1076 construct formed VLPs. We designed another construct (MEG-1076) using the pDEST8-MQG-1076 as template but with a new sense primer and used KOD-plus DNA polymerase according to the manufacture's instructions (Toyobo, Japan). The MEG-1076 construct had the same nucleotide point mutations at positions 1076 in VP1 and 1895 in VP2 as the MQG-1076 construct but not at nucleotide 4 in VP1 (Fig. 1 ). Lastly, we designed a third construct with the native sequence (Wt construct) using KOD-plus DNA polymerase and the original cDNA [ 4 ]. PCR-amplified fragments were cloned into the Gateway Expression System (Invitrogen, Carlsbad, Calif.) as previously described [ 4 ]. The insert sequences of the pDONR8 plasmids were confirmed, including the partial upstream and downstream sequences on the plasmids in which were found to be identical for the Wt and MEG-1076 constructs. Sequencing was performed as previously described [ 18 ]. Expression of rVP1 in insect cells Recombinant bacmids were transfected into Sf9 cells (Riken Cell Bank, Japan) and the recombinant baculoviruses was collected as previously described [ 4 ]. The expression of the rVP1 constructs were analyzed by infecting recombinant baculoviruses at a MOI of 14.5 in 2.7 × 10 6 confluent Tn5 cells in 1.5 ml of Ex-Cell 405 medium followed by incubation at 26°C. The total culture medium was harvested 1, 2, 3, 4, 5, 6, 7, and 8 dpi. The culture medium was centrifuged for 10 min at 3,000 × g , and further centrifuged for 30 min at 10,000 × g . The VLPs in the culture medium were further concentrated by ultracentrifugation for 2 h at 45,000 rpm at 4°C (Beckman TLA-55 rotor), and then resuspended in 30 μl of Grace's medium. The cell lysate from the first centrifuge was resuspended in 200 μl of Grace's medium and stored at 4°C. Northern blotting Total RNA was prepared from the attached cells at 1, 2, 3, 4, 5, and 6 dpi with 1 ml of Isogen (Nippon Gene, Japan). For 7 and 8 dpi, the cell culture medium (containing unattached cells) was collected and centrifuged for 5 min at 3,000 × g , the supernatant removed, and then the cells were dissolved with 1 ml of Isogen. The cells were stored at -80°C. RNA was purified by a chloroform/ ethanol method (Nippon Gene, Japan). Briefly, RNA was mixed with chloroform, centrifuged at 12,000 × g for 15 min at 4°C, and the aqueous layer collected. This was repeated once, and then the aqueous layer collected and mixed with isopropanol and stored overnight at -20°C. The solution was mixed, centrifuged at 12,000 × g for 15 min at 4°C, and the supernatant discarded. The pellet was resuspended in 80% ethanol, centrifuged at 12,000 × g for 15 min at 4°C. This was repeated once, and then the pellet air-dried and resuspended in 25 μl of TE, and stored at -80°C. The amounts of purified RNA were determined spectrophotometrically (Bio-Rad, USA). The same amounts (500 ng) of total RNA were loaded for each construct and each dpi onto a 2% denaturing agarose gel containing formaldehyde. The amounts of total RNA were compared using SYBR Gold staining (Invitrogen, USA). RNA was transferred to a positively charged nylon transfer membrane (Hybond-N+; Amersham Biosciences, Ireland) under vacuum (VacuGene XL; Pharamacia LKB, Sweden) and analyzed by Northern blotting according to the DIG Northern Starter Kit (Roche, USA), except for a minor modification. Briefly, a RNA probe corresponding to Mc114 VP1 position 157 to 1283 (anti-VP1) was generated from a PCR fragment (native sequence) according to the manufacture's instructions (Roche, USA). Hybridization was performed overnight at 68°C with anti-VP1 in 10 ml of ultrasensitive hybridization buffer (Ambion, Canada). After hybridization, immunological detection was performed according to the manufacture's instructions (Roche, USA). Western blotting, ELISA, EM, and protein sequencing Western blotting, ELISA, and EM were used to examine rVP1 expression as previously described [ 4 ]. However, it should be acknowledged that the hyperimmune rabbit and guinea pig antisera were raised against the MQG-1076 VLPs. Protein sequences were determined by an Edman's degradation method. Amino acid alignment VP1 nucleotide sequences were translated using Genetyx software (software development Co. Version 11.2.2) and submitted to online ClustalW at DDBJ . In total, we aligned different 21 SaV GI, GII, GIII, GIV, and GV sequences, and included: Arg39, AY289803; Bristol, AJ249939; C12, AY603425; Cruise ship/00, AY289804; PEC, AF182760; Dresden, AY694184; Hou-7, AF435814; Houston/86/US, U95643; Houston/27/90/US, U95644; London/29845/92/UK, U95645; Lyon/598/97/F, AJ271056; Manchester, X86560; Mc2, AY237419; Mc10, AY237420; Mex340/1990, AF435812; Mex14917/00, AF435813; NK24, AY646856; Parkville, U73124; Potsdam, AAG01042; Plymouth, X86559; Sapporo/82/Japan, U65427; and Sakaeo-15, AY646855. Competing interests The author(s) declare that they have no competing interests. Authors' contributions GH carried out the study and wrote the manuscript. KK, TO, KN, and NT participated in the design of the study and helped to draft the manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC553994.xml
545648	Observed intra-cluster correlation coefficients in a cluster survey sample of patient encounters in general practice in Australia	Background Cluster sample study designs are cost effective, however cluster samples violate the simple random sample assumption of independence of observations. Failure to account for the intra-cluster correlation of observations when sampling through clusters may lead to an under-powered study. Researchers therefore need estimates of intra-cluster correlation for a range of outcomes to calculate sample size. We report intra-cluster correlation coefficients observed within a large-scale cross-sectional study of general practice in Australia, where the general practitioner (GP) was the primary sampling unit and the patient encounter was the unit of inference. Methods Each year the Bettering the Evaluation and Care of Health (BEACH) study recruits a random sample of approximately 1,000 GPs across Australia. Each GP completes details of 100 consecutive patient encounters. Intra-cluster correlation coefficients were estimated for patient demographics, morbidity managed and treatments received. Intra-cluster correlation coefficients were estimated for descriptive outcomes and for associations between outcomes and predictors and were compared across two independent samples of GPs drawn three years apart. Results Between April 1999 and March 2000, a random sample of 1,047 Australian general practitioners recorded details of 104,700 patient encounters. Intra-cluster correlation coefficients for patient demographics ranged from 0.055 for patient sex to 0.451 for language spoken at home. Intra-cluster correlations for morbidity variables ranged from 0.005 for the management of eye problems to 0.059 for management of psychological problems. Intra-cluster correlation for the association between two variables was smaller than the descriptive intra-cluster correlation of each variable. When compared with the April 2002 to March 2003 sample (1,008 GPs) the estimated intra-cluster correlation coefficients were found to be consistent across samples. Conclusions The demonstrated precision and reliability of the estimated intra-cluster correlations indicate that these coefficients will be useful for calculating sample sizes in future general practice surveys that use the GP as the primary sampling unit.	Background Cluster sample study designs are a cost-effective way of sampling difficult to reach populations. Examples include sampling schools to obtain cluster samples of students or medical practitioners to sample patients[ 1 ]. Cluster samples violate the simple random sample assumption of independence of observations, since observations are sampled from within the selected cluster – defined as the primary sampling unit. Observations within a cluster may be more alike than observations across clusters. This intra-cluster correlation leads to increased variation between clusters compared to the variation within clusters. Failure to account for intra-cluster correlation when designing a study where participants are recruited within clusters will lead to an under-powered study. To allow for any loss in power and precision, a cluster sample requires a larger sample size to answer the same research question as a study using simple random sampling [ 2 - 4 ]. Both the size of the intra-cluster correlation and the number of observations sampled within each cluster influence the power of the study. Even for a small intra-cluster correlation, as is often found in general practice and community samples, the loss of power can be appreciable, particularly if the size of the cluster is large[ 1 , 4 ]. Estimates of the size of intra-cluster correlations come from post hoc examination of studies that have used either allocation or sampling by cluster and a number of intervention studies have published observed intra-cluster correlation coefficients [ 4 - 6 ]. Many intervention studies however, still fail to report intra-cluster correlation coefficients[ 7 ] and there is even less information reported on survey studies that employ a cluster sample[ 1 , 8 ]. The lack of published estimated intra-cluster correlations continues to hamper the design of studies that employ a cluster sample[ 9 ]. Intra-cluster correlation varies within a study and depends on the outcome under analysis[ 1 , 4 , 6 ]. The intra-cluster correlation of the same outcome may also vary across studies depending on the primary sampling unit, and whether outcomes are reported as prevalence rates or modeled in association with other variables[ 1 , 4 ]. Researchers need reliable estimates of the intra-cluster correlations, specific to the primary sampling unit and selected outcomes of interest when making sample size calculations. These estimates will assist in deciding the trade off between cluster number and cluster sub-sample size in a study design[ 10 ]. There is however, little published on the estimated intra-cluster correlation coefficients in the Australian context, especially for primary health surveys where the health practitioner is the primary sampling unit. Research questions In one Australian study Carlin and Hocking[ 1 ] examined the intra-cluster correlation in two cross-sectional cluster surveys of school children that used the school as the primary sampling unit. The researchers observed that design effects for sociodemographic variables were larger than for morbidity related variables. Furthermore intra-cluster correlation was greater for descriptive outcomes such as prevalence estimates, means and proportions than for measures of association between variables such as regression coefficients and odds ratios. We wanted to examine whether these patterns could be generalised to other large cluster survey studies in the primary care setting. This paper reports some of the intra-cluster correlations observed in the Bettering the Evaluation and Care of Health (BEACH) program, a large cross-sectional survey of general practice patient encounters in Australia, where a random sample of general practitioners was used as the primary sampling unit. The BEACH study draws a new random sample of Australian general practitioners (GPs) each year, and this provided an opportunity to assess the stability of intra-cluster correlation coefficients across successive samples. If a population is re-sampled using the same cluster survey design, will the intra-cluster correlation coefficient for a particular outcome be the same across samples? This analysis takes an applied approach, examining the observed intra-cluster correlations for a range of demographic, morbidity and treatment outcomes. Methods The BEACH program is a continuous study of general practice activity commenced in 1998. The BEACH method is described in detail elsewhere and a brief summary is reported below[ 11 ]. Cluster sample design A random sample of approximately 1,000 general practitioners (GPs) is drawn each year from the Health Insurance Commission's sampling frame of the population of GPs in Australia. The GP population is randomly ordered into a list and GPs are recruited sequentially from the list, with re-randomisation of the sampling frame every three years[ 11 , 12 ]. GPs are sampled without replacement and have one chance of selection over three years. Sampling is continuous across the year, with around 20 GPs participating in the study in any one week. Each GP completes details of 100 consecutive patient encounters. The GP is the primary sampling unit (PSU), while the primary unit of inference is the patient encounter. Data elements A single page encounter form contains elements including: • Patient age and sex. • Whether English was the main language spoken at home. • Whether the patient holds an Australian health care concession card. • The problems managed by the GP at the encounter (up to four problems per encounter). • Treatments received at the encounter, including medications, other procedures, referrals and orders for pathology and imaging tests. Although sample weights are calculated each year for population estimates[ 13 ], the outcomes reported in this paper are unweighted to allow us to calculate estimates of the intra-cluster correlation based on the observed variance in the sample data. Descriptive outcomes Descriptive outcomes were defined as rates, means and percentages of single variables, e.g. mean age, per cent of encounters with female patients, per cent of encounters where at least one respiratory problem was managed. BEACH samples the GP-patient encounter, not independent patients. If a patient returns to the GP in the sampling period then that patient contributes two (or more) encounters to the sample. Therefore BEACH estimates are not true "prevalence" rates because the denominator, the population of GP-patient encounters, is many times larger than the population of all general practice patients. To avoid misunderstanding in this paper we have used the term "descriptive" rather than "prevalence" to report single variable estimates and their accompanying intra-cluster correlation coefficients. Descriptive rates are interpreted for example as "Proportion of patients at encounter who are female". Demographic variables Demographic variables include patient sex, patient age, whether the patient held a health care concession card and whether the main language spoken at home was not English. Morbidity variables Problems were classified using the International Classification of Primary Care (ICPC-2)[ 14 ]. The upper level of ICPC-2 classifies problems according to the body system involved, for example skin problems, respiratory problems, cardiovascular problems and problems of the digestive system etc. There are an additional three chapters for psychological problems, social problems and problems of a general or unspecified nature. Morbidity estimates are expressed as the percent of patient encounters where at least one problem from the chapter was managed. The total number of problems managed by the GP at the encounter was also included as an outcome. Treatment outcomes Treatment outcomes included the proportion of encounters that resulted in at least one medication, the proportion that received at least one referral, the proportion that received at least one order for an imaging test and the proportion receiving at least one order for a pathology test. Association outcomes Intra-cluster correlation coefficients were calculated for associations between variables using logistic regression e.g.: the effect of patient age (predictor) on the rate of cardiovascular problems (outcome). Design effect Obtaining the sample size for cluster designs involves calculating the sample size under the assumption of simple random sampling and then inflating the number of observations to allow for the design effect of the cluster sample. The design effect (Deff) of an outcome has been defined as the ratio of the variance taking into account the cluster sample design and the variance of a simple random sample (srs) design with the same number of observations[ 1 ]. Deff = Variance (clustersample) /Variance (srs) Intra-cluster correlations and their standard errors for the outcome variables were calculated using the method described by Carlin & Hocking[ 1 ]. Specifically STATA 7 was used to calculate the design effects using the "survey estimator" procedures, which were purposefully designed to analyse complex survey data. STATA 7 calculates the design effect directly from the ratio of the estimated variances[ 15 ]. The intra-cluster correlation coefficient (ICC) was then calculated from the design effect using the formula: ICC = ( Deff - 1)/( k - 1) and the approximate standard error (SE) of the intra-cluster correlation was calculated using the formula[ 1 , 6 ]: where m = number of clusters, k = mean number of observations per cluster. The intra-cluster correlations and respective 95% confidence intervals for the second BEACH year sample from the period April 1999 to March 2000 were compared against those in the year 5 sample (April 2002 to March 2003) to assess whether the intra-cluster correlations were consistent across samples over time. All calculations specified the GP as the primary sampling unit. Results From April 1999 to March 2000, 1,047 GPs were recruited, recording a sample of 104,700 patient encounters. From April 2002 to March 2003, 1,008 GPs were recruited and 100,800 encounters recorded. Table 1 shows the age and sex distribution of the two samples of GPs compared with the sampling frame of the population of Australian GPs in the year April 2002 to March 2003[ 13 ]. The two GP samples were comparable to the GP population in terms of distribution by age, sex and state. Table 1 Comparison of GP participants and all active recognised Australian GPs. BEACH April 1999–March 2000 % (95%CI) (N = 1,047) BEACH April 02–March 03 % (95%CI (N = 1,008) Australian GPs April 02 to March 03 % (N = 17,884)[13] Males 69.6 (66.8,72.4) 64.8 (61.8,67.7) 66.8 Age group <35 8.4 (6.7,10.1) 7.3 (5.7,9.0) 9.7 35–44 32.4 (29.6,35.3) 26.6 (23.9,29.3) 25.1 45–54 32.4 (29.6,35.3) 35.2 (32.3,38.2) 33.1 55+ 26.7 (24.1,29.4) 30.9 (28.0,33.7) 32.0 State NSW 37.4 (34.5,40.4) 39.6 (36.7,42.7) 33.6 Victoria 20.1 (17.7,22.5) 18.8 (16.4,21.3) 24.5 Queensland 20.2 (17.8,22.6) 21.2 (18.7,23.8) 18.5 South Australia 9.1 (7.3,10.8) 6.2 (4.7,7.6) 8.7 Western Australia 8.8 (7.1,10.5) 8.9 (7.2,10.7) 9.5 Tasmania 2.4 (1.5,3.3) 2.8 (1.8,3.8) 2.9 ACT 1.1 (0.5,1.8) 1.4 (0.6,2.0) 1.5 NT 0.9 (0.3,1.4) 1.1(0.4,1.7) 0.8 The two samples of patient encounters were similar in terms of demographics (Table 2 ) In the year 1999–00, 59.0% of encounters were with female patients compared with 59.3% in 2002–03. The samples were comparable in terms of the mean age of patients, the proportion of health care card holders, and encounters with patients from a non-English speaking background. Table 2 Descriptive parameters of demographic, morbidity and treatment variables with design effects (Deff), intra-cluster correlation coefficients (ICC) and standard errors of ICC (SE) for sample year April 1999 to March 2000 (N = 1,047 general practitioners): compared with ICC and SE for sample April 2002 to March 2003 (N = 1,008 GPs). 1999–2000 (N = 1,047 GPs) 2002–2003 (1,008 GPs) Parameter Estimate(SE) Deff ( a ) ICC(SE) Estimate ICC(SE) Demographics Sex (% female) 59.0 (.39) 6.4 .055 (.003) 59.3 .066 (.003) Age (years) – mean 44.5 (.31) 16.6 .159 (.006) 45.4 .153 (.006) Holds health care card (%) 40.1 (.70) 21.4 .206 (.007) 42.7 .209 (.008) Patient language ( c ) (%) 7.0 (.53) 45.6 .451 (.011) 8.8 .423 (.011) Morbidity Number of problems (per 100 encounters) 149.5 (.86) 13.6 .127 (.005) 148.7 .141 (.006) Problem by ICPC-2 chapter (b) Cardiovascular (%) 15.2 (.29) 6.7 .057 (.003) 15.3 .056 (.003) Respiratory (%) 21.0 (.26) 4.2 .032 (.002) 19.0 .040 (.002) Psychological (%) 10.6 (.25) 6.8 .059 (.003) 10.6 .061 (.003) Endocrine/Metabolic (%) 8.8 (.18) 4.2 .032 (.002) 10.1 .031 (.002) Blood (%) 1.7 (.08) 3.7 .027 (.002) 1.4 .007 (.001) Digestive (%) 9.6 (.12) 1.8 .008 (.001) 9.7 .010 (.001) Eye (%) 2.8 (.06) 1.5 .005 (.001) 2.6 .003 (.001) Musculoskeletal (%) 16.3 (.23) 4.1 .032 (.002) 16.5 .045 (.002) Skin (%) 16.1 (.19) 2.7 .017 (.001) 15.9 .042 (.002) General unspecified (%) 14.0 (.22) 4.3 .034 (.002) 15.8 .043 (.002) Treatment (% of encounters) Any medications 67.0 (.37) 6.6 .056 (.003) 64.4 .068 (.003) Any referrals 11.2 (.20) 4.1 .031 (.002) 12.0 .033 (.002) Any pathology tests ordered 14.7 (.26) 5.8 .048 (.002) 16.0 .046 (.002) Any imaging tests ordered 6.9 (.15) 3.8 .028 (.002) 7.8 .029 (.002) ( a ) Average number of observations per cluster k = 100, except age ( k = 99.2) and sex ( k = 98.8). ( b ) Per cent of encounters where at least one problem from the chapter was managed. ( c ) Patient speaks a language other than English at home. Descriptive ICCs (March 1999–April 2000) Demographics For descriptive estimates of demographic variables the intra-cluster correlation ranged from 0.055 for sex of patient at encounter to 0.451 for language spoken by the patient at home. (Table 2 ). With a standard cluster size of 100 encounters this produced design effects ranging from 6.4 for patient sex to 45.6 for non-English speaking background. Morbidity (ICPC body chapter) For descriptive estimates of the management rates of morbidity problems, the intra-cluster correlations ranged from 0.005 for estimates of eye problems to 0.059 for estimates of psychological problems, with design effects of between 1.5 and 6.8 respectively. Treatments The intra-cluster correlation coefficients for treatments received ranged from 0.028 for any imaging tests ordered to 0.056 for any medications. Association ICCs For bivariate relationships between an outcome and predictor, the association ICCs were considerably smaller than the descriptive ICCs (Table 3 ). This pattern was observed for both demographic and morbidity outcomes. When analysing the association between holding a health care card and other demographic variables, the ICCs ranged from 0.012 for patient sex to 0.128 for language background (Table 3 ), which were smaller than for the descriptive estimate of the percentage holding a health care card (Table 2 ). Table 3 Associations between demographic and morbidity variables, measured as odds ratios, with design effect (Deff) and intra-cluster correlation coefficients (ICC) with standard errors (SE) for sample year April 1999 to March 2000 (N = 1,047 general practitioners): and ICC and SE for sample April 2002 to March 2003 (N = 1,008 GPs). 1999–2000 (N = 1,047 GPs) 2002–2003 (1,008 GPs) Outcome* Predictor Odds Ratio Deff ( a ) ICC (SE) ICC (SE) a) Demographic Patient holds health care card Female patient 1.07 2.2 .012 (.001) .018 (.001) Age (years) 1.03 6.5 .056 (.003) .073 (.003) Patient language ( b ) 1.26 13.7 .128 (.005) .114 (.005) Patient language ( b ) Female patient 0.94 3.7 .028 (.002) .024(.001) Age (years) 1.00 11.2 .104 (.004) .098 (.004) b) Morbidity Chapters Cardiovascular Female patient 0.90 1.3 .003 (.001) .004 (.001) Age (years) 1.05 2.1 .011 (.001) .017 (.001) Holds health care card 2.55 2.4 .014 (.001) .018 (.001) Patient language ( b ) 1.17 5.2 .042 (.002) .034 (.002) Respiratory Female patient .86 1.3 .003 (.001) .003 (.001) Age (years) .99 2.5 .015 (.001) .022 (.001) Holds health care card .89 2.1 .011 (.001) .017 (.001) Patient language ( b ) 1.17 2.9 .020 (.001) .025 (.002) Psychological Female patient 1.13 2.2 .013 (.001) .008 (.001) Age 1.01 3.3 .024 (.001) .022 (.001) Holds health care card 1.89 2.4 .014 (.001) .020 (.001) Patient language ( b ) .74 5.0 .040 (.002) .026 (.002) Endocrine/metabolic Female patient .95 1.6 .006 (.001) .004 (.001) Age 1.03 2.1 .011 (.001) .012 (.001) Holds health care card 1.63 2.4 .014 (.001) .009 (.001) Patient language ( b ) 1.58 3.3 .023 (.001) .019 (.001) * Each predictor is fitted alone, each line represents a separate model. ( a ) Average number of observations per cluster k = 100, except age ( k = 99.2) and sex ( k = 98.8). ( b ) Patient speaks a language other than English at home. When analysing the association between cardiovascular problems as the outcome and selected demographic variables, the ICCs ranged from 0.042 (patient language as the predictor) to 0.003 (patient sex as the predictor)(Table 3 ) compared with the larger ICC of 0.057 when describing the rate of cardiovascular problems (Table 2 ). Comparison of year 2 (April 1999 to March 2000) and year 5 (April 2000 to March 2003) For descriptive outcomes the intra-cluster correlations for year 2 and year 5 samples there was consistency in the patterns of ICCs across samples. (Table 2 and Figure 1 ). One exception was for the management of problems related to the blood system, where the descriptive ICC in 1999–00 was 0.027 (95% CI: 0.024–0.030), three times that observed in 2002–03 (0.007, 95% CI: 0.005–0.008). This was influenced by one GP in the 1999–00 sample who managed blood-related problems at more than 50% of encounters. When this GP was removed, the descriptive ICC for blood related problems in 1999–00 was 0.011 (95%CI: 0.009–0.013), much closer to the ICC observed in 2002–03. Figure 1 Intra-cluster correlation(ICC) and 95% confidence intervals for descriptive and morbidity outcomes in two BEACH samples, April 1999–March 2000 (N = 1047 GPs) and April 2002–March 2003(N = 1008 GPs) * Total problems = the number of problems managed at the current encounter. The intra-cluster correlation for associations between morbidity outcomes and demographic predictors are shown in Table 3 and Figure 2 . Although the intra-cluster correlations for associations between variables across each year were statistically significantly different for some outcomes, in these instances the ICCs were very small and the difference between samples was less than 0.01. Figure 2 Intra-cluster correlation (ICC) and 95% confidence interval for association between morbidity outcomes with health care card status as predictor in two BEACH samples, April 1999–March 2000 (N = 1,047 GPs) and April 2002–March 2003 (N = 1,008 GPs) Discussion The pattern of intra-cluster correlation and design effects observed in the BEACH study agree with Carlin and Hocking's observations in other cluster sample surveys[ 1 ]. Generally we found that sociodemographic variables had larger intra-cluster correlation coefficients than morbidity or treatment variables and outcomes fitted with explanatory variables had smaller intra-cluster correlation coefficients than outcomes reported as descriptive rates. Therefore when designing cluster sample surveys, the effect of the intra-cluster correlation on power calculations, depends on whether the main outcomes of interest are demographic or morbidity variables, and whether the main aims of the study are descriptive or predictive[ 1 ]. We further demonstrated that for a large range of variables the size and patterns of intra-cluster correlation coefficients for particular outcomes were mostly consistent over different sample periods. This indicates that intra-cluster correlation is quite stable when re-sampling a population using the same primary sampling unit, where the number of clusters is sufficiently large. This repeatability demonstrates the validity of using published intra-cluster correlation coefficients to predict intra-cluster correlation in future studies of similar design. Precision can be an issue for estimating intra-cluster correlation, especially for studies with a small number of clusters[ 10 ]. The large number of clusters in this study gave good precision in the estimated intra-cluster correlation coefficients[ 10 ]. There are no other published studies in general practice in Australia with such a large sample of clusters and a large balanced sample of observations per cluster, thus estimating intra-cluster correlation with a high degree of precision. The BEACH study also has the advantage of being a nationwide survey of general practice where the generalisability to Australian general practice has been well-described[ 11 ]. Most research in primary care in Australia is done through general practice, so estimating the intra-cluster correlation for a range of outcomes is important for future researchers who intend to use the GP as the primary sampling unit. The good representation of general practice in the BEACH study, the large sample of clusters and the large cluster size, allow the intra-cluster correlation coefficients reported here to be generalisable to other general practice surveys. These reported intra-cluster correlation coefficients are also likely to be useful for intervention studies that use the GP as the unit of randomisation[ 1 ]. Treatments received at the encounter are outcomes that arise as a result of the GP-patient interaction. Treatments are directly related to GPs' behaviour and so might be expected to be highly correlated within clusters. However we found that the intra-cluster correlation coefficients for medications, referrals, imaging and pathology orders were of a similar order to those for health problems managed. The difference across samples in the intra-cluster correlation coefficients for the management of blood system problems indicates that, even in large samples, intra-cluster correlation may be influenced by GPs in the sample who specialise in particular areas of health. Demographic variables are collected in the BEACH study for the purpose of understanding health status and health service use and these variables are likely to be correlated to a patient's choice of GP. Furthermore a patient can be sampled more than once if they return to the GP during the survey period. Therefore the intra-cluster correlation estimated for demographic variables may be larger than those that have been reported in community based surveys[ 1 , 8 ]. Conclusions As with cluster randomised trials, researchers in primary health care need access to a range of estimates of intra-cluster correlation for the successful planning of cluster survey study designs. We have reported relatively stable intra-cluster correlation coefficients for a range of outcomes across two independent random samples in a large-scale representative survey of general practice in Australia. The demonstrated precision and reliability of the estimated intra-cluster correlations indicate that these coefficients will be useful for calculating sample sizes in future general practice surveys that use the GP as the primary sampling unit. Abbreviations GP: General Practitioner ICC: Intra-cluster correlation coefficient Competing interests The author(s) declare that they have no competing interests. Authors contributions SK conceived the research questions, undertook the analysis and wrote the main draft of the manuscript. PC participated in formulating the research questions and the design of the analysis, undertook a literature search and assisted in the writing of the main draft and subsequent revisions of the manuscript. Pre-publication history The pre-publication history for this paper can be accessed here:	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC545648.xml
553980	The Arabidopsis Mei2 homologue AML1 binds AtRaptor1B, the plant homologue of a major regulator of eukaryotic cell growth	Background TOR, the t arget o f the antibiotic r apamycin in both yeast and mammalian cells, is a potent cell growth regulator in all eukaryotes. It acts through the phosphorylation of downstream effectors that are recruited to it by the binding partner Raptor. In Arabidopsis, Raptor activity is essential for postembryonic growth. Though comparative studies suggest potential downstream effectors, no Raptor binding partners have been described in plants. Results AtRaptor1B, a plant Raptor homologue, binds the AML1 ( A rabidopsis M ei2- l ike 1) protein in a yeast two-hybrid assay. This interaction is mediated by the N-terminal 219 residues of AML1, and marks AML1 as a candidate AtTOR kinase substrate in plants. The AML1 N-terminus additionally carries transcriptional activation domain activity. Plants homozygous for insertion alleles at the AML1 locus, as well as plants homozygous for insertion alleles at all five loci in the AML gene family, bolt earlier than wild-type plants. Conclusion AML1 interacts with AtRaptor1B, homologue of a protein that recruits substrates for phosphorylation by the major cell-growth regulator TOR. Identification of AML1 as a putative downstream effector of TOR gives valuable insights into the plant-specific mode of action of this critical growth regulator.	Background TOR, the t arget o f the antibiotic r apamycin in both yeast and mammalian cells, is a major regulator of cell growth and translation [ 1 ]. TOR is a large (over 2,400 residues) protein kinase [ 2 ] present in all eukaryotes analyzed. It is thought to act in a nutrient-sensitive complex TORC1 with Raptor ( r egulatory a ssociated p rotein of TOR) and another protein to regulate cell growth [ 3 - 7 ] – though there is some debate on the nutrient sensitivity of the complex [ 8 ]. Raptor, a protein with HEAT and WD-40 protein interaction domains, recruits substrates for phosphorylation by TOR in yeast and mammals [ 3 , 7 ]. TOR also acts in a second, nutrient-insensitive complex without Raptor to regulate the cytoskeleton [ 3 , 8 , 10 ]. Disruption of the Arabidopsis TOR homologue AtTOR is lethal early in plant embryonic development [ 11 ]. Disruption of AtRaptor (encoded by two paralogous loci in Arabidopsis) causes seedling developmental arrest but allows normal embryonic development [[ 12 ], manuscript in preparation], suggesting that AtTOR embryonic activity is independent of AtRaptor and that the TOR-Raptor complex has been adapted in the ancestor of the angiosperms to regulate post-embryonic growth. In support of this, AtTOR has been show to be expressed in dividing and expanding cells [ 11 ]. Thus, identifying downstream activators of TOR signaling may provide insights into the activity of AtTOR in post-embryonic growth. Mei2 is a putative TOR substrate and potent meiosis-signaling molecule identified in the fission yeast Schizosaccharomyces pombe [ 13 , 14 ]. Mei2 is bound by the Raptor homologue Mip1 [ 15 ], and is an inactive phosphoprotein under high nutrient conditions [ 16 , 17 ] – conditions which promote TOR kinase activity. The kinase governing two of the three Mei2 phosphorylation sites is known [ 16 ]; TOR is a strong candidate for the kinase governing the third. In diploids under low nutrient conditions, unphosphorylated Mei2 accumulates and localizes to the nucleus, where it binds to a noncoding, mRNA-like RNA molecule meiRNA in an interaction mediated by the third of its three RNA Recognition Motifs (RRMs)[ 14 , 16 , 18 ]. The Mei2-meiRNA interaction occurs as meiRNA is being transcribed, tethering Mei2 to the meiRNA locus [ 19 ]. Accumulation of Mei2 at this focused point immediately precedes meiosis. Mei2-like proteins are predicted in a wide range of organisms [ 20 , 21 ] including some fungi, alveolates, a diatom, and all land plants, but they are absent from metazoans and budding yeast. In land plants, predicted Mei2-like proteins form a small conserved gene family, many of whose members' transcripts accumulate in the shoot apical meristem specifically or in the shoot apical meristem in addition to a range of mature tissues [ 20 ]. A rabidopsis M ei2- l ike 1 (AML1), the first member of this family to be described, was isolated in a screen for plant cDNAs whose expression could complement defects in the fission yeast meiosis signaling pathway [ 22 ]. Like Mei2, AML1 has three RRMs. Expression of a protein fragment containing only the third AML1 RRM was sufficient for restoration of meiosis signaling in fission yeast lines with defects upstream of Mei2, but not in lesions of the mei2 locus itself. T erminal E ar 1 (te1), a more divergent member of the mei2 -like family of genes, regulates leaf initiation in maize; the tassel of mutant plants is encased a whorl of leaves superficially resembling a maize ear [ 23 ]. Given the potential role of downstream AtTOR effectors in post-embryonic growth, the intriguing signaling activity of Mei2 in fission yeast, and the known interaction between Mei2 and the fission yeast Raptor protein, we asked if the Mei2-Raptor interaction was conserved between its Arabidopsis orthologues AML1 and AtRaptor1B. Results AML1 and AtRaptor1B interact in a yeast two-hybrid assay To test for an interaction between the Arabidopsis proteins AML1 and AtRaptor1B, the open reading frames (ORFs) of each of the transcripts predicted to encode these proteins were amplified via PCR with primers carrying suitable restriction sites for cloning into the yeast two hybrid vectors pGADT7 and pGBKT7. pGADT7 encodes a transcriptional activation domain (AD) which can recruit the yeast transcription machinery. pGBKT7 encodes a DNA binding-domain (BD) which binds to the promoters of ADE2 and HIS3 . pGADT7 and pGBKT7 additionally carry the genes LEU2 and TRP1 . When transformed into the yeast line AH109, which is leu2 - trp1 - ade2 - his3 - , co-transformed cells will grow on yeast medium lacking leucine and tryptophan. Cells co-transformed with two-hybrid constructs encoding proteins that interact will grow on media additionally lacking histidine and adenine (selective media). AH109 yeast cells co-transformed with clones encoding AML1 and AtRaptor1B in complementary two-hybrid vectors grew on selective media. Co-transformations of control pGBKT7 and pGADT7 empty vectors, or either control vector co-transformed with its complement harboring the AML1 or AtRaptor1B ORF, yielded cells able to grow on media lacking leucine and tryptophan but not on selective media. Evidence from other systems indicates that Raptor protein fragments lose the ability to bind substrates [ 4 ]. Therefore, only full length AtRaptor1B was tested in this assay. Mei2, the fission yeast AML1 homologue, is highly modular. It is divided into distinct N-terminal and C-terminal domains. The N-terminal half of the protein appears to play a regulatory role [ 24 ]. The C-terminal half of Mei2 is sufficient to complement lesions in the mei2 locus [ 16 ]. Additionally, the AML1 C-terminal half, expressed in fission yeast meiosis signaling mutants, is able to complement meiosis signaling defects upstream of Mei2 [ 22 ]. Therefore we generated clones encoding fragments of AML1 and assayed them for interactions with full length AtRaptor1B. AML1 fragments N412 and N219, comprised of the first 412 or the first 219 residues of the 915 residue AML1 protein, restored growth on media lacking leucine, tryptophan, histidine and adenine when cloned into pGADT7 and co-transformed with pGBK-Raptor (Fig. 1 ). Neither N163, nor 155–219, (the fragments which together comprise N219), nor 155–412 (which with N163 comprises N412), could restore growth when cloned into pGADT7 and co-transformed with pGBK-Raptor. AML1 fragment 402C, comprised of residues 402 to the C-terminus of the protein, and all C-terminal fragments tested (695C, 402–704) failed to restore growth on selective medium. None of the AML1 fragments cloned into pGADT7, and cotransformed with empty pGBKT7 control vector, yielded transformants able to grow on selective medium. AML1 fragments N412 and N163 harbor activation domain activity in a yeast one-hybrid assay AML1 N412, cloned into pGBKT7, restored growth on selective medium to cells co-transformed with either pGAD:Raptor or pGADT7. To investigate this result further, we tested all AML1 fragments for native transcriptional activation domain activity in a yeast one-hybrid assay. AML1 fragments were cloned into pGBKT7, singly transformed into AH109 yeast cells, and assayed for growth on media lacking tryptophan (to confirm transformation) and media additionally lacking adenine. The DNA binding-domain of the pGBKT7 tethers any C-terminally fused fragments to the ADE2 promoter. In the absence of a binding partner, BD-fusion chimeric proteins trigger transcription of ADE2 only if the protein fused to the BD contains native transcriptional activation activity. AML1 N412 and AML1 N163, but not full length AML1, AML1 N219 or any C-terminal AML1 fragments, were able to restore growth on media lacking tryptophan and adenine (Fig. 2 ). AML1 N122 was similarly unable to restore growth. AML1 fragments 155–219 and 155–412, in pGBKT7, could not be stably transformed into yeast. These results were observed in multiple independent transformant lines for a given construct. Cotransformation of any of the pGBKT7-derived constructs with pGADT7 did not affect the growth of any of the transformed lines on media lacking tryptophan and adenine. Plants homozygous for insertion alleles of AML1 and of all five AML family members show early flowering An interaction with AtRaptor1B points to AML1 as a downstream effector of TOR signaling in plants. Additionally, the dramatic phenotype mei2 disruption and the intriguing mode of Mei2 action led us to ask what the consequences would be of disruption of the AML1 locus and of all five AML gene family members. To obtain insertion alleles in AML1 and other AML gene family members, we screened the insertion allele populations at the University of Wisconsin Arabidopsis Knockout facility [ 25 ], obtaining alleles harboring insertions in AML1 , AML3 , AML4 and AML5 (Fig. 3 ). An AML2 insertion allele was obtained from the SIGnAL collection at the Salk Institute [ 26 ]. By RT-PCR using primers which anneal to the cDNA at sites spanning the insertion site of each insertion allele, we established that no wild-type transcripts accumulate in homozygous mutants (Fig. 4 ). By a series of crosses, we then generated higher-order insertion homozygotes, culminating in the quintuple insertion homozygotes Q6 and Q17. All lines were viable and fertile. AML insertion homozygotes bolted earlier than wild-type lines (Fig. 5A and 5B ). This effect was independent of the number of insertion alleles carried by the mutants; AML 'quint' lines Q6 and Q17 were not qualitatively different than lower order insertion allele homozygotes. Additionally, AML insertion homozygote seedlings were assayed for a differential response to a range of signaling molecules. Seedlings were germinated on culture medium supplemented with the gibberellic acid GA 3 , paclobutrazol, the auxin 2,4-D, 1-amino-cyclopropane-1-carboxylic acid, 1% sucrose, 6% sucrose and kinetin, and in the dark. Quintuple insertion mutants responded slightly more than wild-type seedlings to GA 3 as measured by the change in seedling length in the presence vs. in the absence of the hormone. This effect was repeatable but weak, and no other differential hormonal response was observed (data not shown). Sequence downstream of the mutant allele insertion sites is transcribed Given the mild phenotype of the AML insertion homozygotes, we further investigated the extent of the effect of the insertions at each locus. RT-PCR, as previously stated, showed that no wild-type transcripts accumulate in lines Q6 and Q17. In fission yeast, however, the C-terminal half of the protein is sufficient to complement lesions of the mei2 locus [ 16 ]. The insertion alleles of all but AML4 are disrupted at or near the 5' ends of their predicted coding regions. We therefore designed primers that anneal to the region downstream of the insertion site in each allele and performed polymerase chain reactions to assay for accumulation of fragments capable of encoding the C-terminal half of any of the AML proteins (Fig. 4 ). To determine whether the amplified fragments corresponded to AML cDNAs, we performed restriction digests on the PCR products, which confirmed that the cDNAs originated from AML gene transcripts. Weak amplification of cDNA representing transcripts originating downstream of the insertion site was observed for AML1, AML2 and AML3; amplification of the AML5 3' region was indistinguishable from the amplification seen from of wild-type cDNA template. Transgenic lines overexpressing AML1:GFP or GFP:AML1 fusion proteins could not be recovered The AML1 ORF was separately cloned into the pCambia1302 35S::GFP plant transformation vector both 5' and 3' of the GFP ORF, and the construct was transformed into Arabidopsis via Agrobacterium -mediated floral dip [ 27 ]. Transformants, identified by resistance to the antibiotic hygromycin and confirmed through PCR, were recovered at a very low rate of less than .01%. No fluorescence was observed in any tissues of any transformants assayed, and AML1:GFP transcripts could not be detected via RT-PCR performed on cDNA transcribed from RNA extracted from bulk shoot tissue (data not shown). Discussion Raptor proteins in yeast and mammals function by recruiting substrates for TOR, a central regulator of cell growth in response to nutrients [ 4 , 7 , 28 ]. An interaction with Raptor therefore strongly suggests that a given protein is a TOR substrate and downstream effector of TOR signaling. Plants homozygous for lesions at both AtRaptor loci show normal embryonic development but are unable to maintain shoot meristem activity [ 12 ]. TOR substrates, then, may play a role in regulating meristem-driven post-embryonic growth. The interaction between AML1 and AtRaptor1B implicates the AML family of proteins in TOR signaling. It points specifically to a role for the AML proteins in regulation of shoot meristem activity. An interaction between Mei2 and the fission yeast Raptor homologue Mip1 has been reported previously; indeed, Mip1 ( M ei2 i nteracting p rotein 1) was the first Raptor homologue characterized in any eukaryote [ 15 ]. The conservation of this interaction from fission yeast to plants suggests that the well-characterized Mei2 signaling pathway may provide insight into the function of the AMLs. Mei2 is a potent meiosis-signaling molecule. It triggers pre-meiotic cell differentiation and meiosis in response to nutrient stress [ 13 , 14 ]. Meiosis signaling in fission yeast is a model for cell differentiation in response to external nutrient cues. Thus the AMLs may also play a role in cellular differentiation or in meiosis signaling. Aside from the effect of Mei2 in development, there is the intriguing issue of its mode of action. Mei2 sub-cellular localization is mediated by an interaction with a noncoding, mRNA-like molecule [ 14 , 19 ]. There is a fairly large population of mRNA-like transcripts conserved among land plants despite lacking large conserved open reading frames [ 29 ]. Of these, the conserved alfalfa transcript ENOD40 has been shown to mediate the sub-cellular localization of an RNA-binding protein [ 30 ] and to mediate phytohormone responses [ 31 ]. AML1 may be a binding partner of one or more of these mRNA-like noncoding molecules in plants. The transcriptional activation activity of the AML1 N-terminus observed in the yeast one hybrid assay has not been ascribed to Mei2 and may represent a novel activity of plant Mei2-like proteins. This activation activity localizes to the N-terminal 163 residues, but is strongly influenced by the adjacent residues. Activity is lost in N219, regained on N412 and lost again in full length AML1. This suggests that the AML1 N-terminal half has multiple configurations, and that the accessibility of the activation domain varies among configurations. We should emphasize, however, that the AML1 N-terminal fragment transcriptional activation activity has yet to be shown in planta , and that the activity may result from fusion to the DNA binding domain rather than being present in native AML1. AtRaptor1B binding to AML1 is also localized to the N-terminus, and appears to be mediated by multiple sites in this region. This suggests that the N-terminus may contain a TOR phosphorylation site, and that this site may influence the configuration of the N-terminus. The repeated failure to recover transgenic lines expressing AML1 suggests that its unregulated overexpression is lethal. Future efforts to characterize the AML proteins in transgenic plants may benefit from the use of inducible promoters driving transgene expression to circumvent the putative lethality of unregulated AML expression. Disruption of any of the AML loci causes early bolting in plants grown under long days. However, lines homozygous for insertions in all five AML loci did not differ dramatically from lower-order insertion homozygotes, despite the fact that RT-PCR performed with primers spanning the insertion sites show that the wild-type transcript does not accumulate. Transcripts originating downstream of the insertion sites but still capable of encoding the C-terminal half of the wild-type protein accumulate from all loci but AML4 . This raises the possibility that the AML quintuple insertion homozygote lines do not represent total disruption of AML activity. Four of the five AML open reading frames in the insertion mutant are apparently truncated and all are divorced from their native promoters, but some promoter activity (perhaps from the 35S viral promoters harbored within the inserted DNA) remains and may be sufficient to cause transcription of AML coding region DNA downstream of the insertion site. Viewed in this light, the early flowering phenotype of AML insertion homozygotes may arise not from the total disruption of AML activity but from the accumulation of AML proteins which, due to the truncations in their N-termini caused by the insertions, are no longer bound by AtRaptor1B, no longer phosphorylated by AtTOR, or no longer able to activate transcription of floral repressors. Finally, these results provide a cautionary tale. RT-PCR performed using primers which span an insertion site may not be sufficient to conclude that all activity of a protein of interest is abolished. Conclusion TOR is a major regulator of cell growth in eukaryotes, but little is known about its downstream effectors in plants. This work shows that AML1 binds AtRaptor1B, and suggests that the AML protein family may be phosphorylated by AtTOR in an AtRaptor1B-mediated interaction. The interaction with AtRaptor1B implicates AML1 as a downstream effector of AtTOR kinase signaling, and provides insight into the mode of action of this critical growth regulator. Methods Generating the two-hybrid constructs AML1 was cloned via RT-PCR. The cDNA template was reverse-transcribed using Omniscript (Qiagen) from RNA extracted from bulk shoot tissue using Trizol Reagent (Invitrogen). Restriction sites Nco I and Xma I/ Sma I were added to the 5' and 3' ends of the ORF and of all smaller AML1 fragments via PCR using ExTaq high-fidelity polymerase (Takara). An EST clone (RZL03b06) tagging AtRaptor1B was obtained from Kazusa DNA institute and sequenced. Restriction sites Nco I and Eco RI were added at the 5' and 3' ends of the ORF via PCR. pGBKT7 and pGADT7 are distributed by BDBiosciences. Yeast two-hybrid assay AH109 cells ( leu2 - trp1 - ade2 - his3 - ) were grown in YEPDA liquid plates or on YEPDA plates with 17 g/L Agar-Y (Bio101 Systems). Cells were transformed using the Yeastmaker Yeast Transformation System2 (BD Biosciences) and plated on medium lacking the appropriate macronutrients (Bio101 Systems). Colonies were observed 3–7 days after transformation. Genotyping of insertion alleles DNA from lines harboring insertion alleles was extracted using the alkaline boiling method [ 32 ]. Provisional homozygotes were confirmed via a second extraction using the C-TAB DNA extraction protocol. PCR to assay for wild-type and insertion alleles was performed in 20 uL volumes using ExTaq polymerase and buffers and the following cycling parameters: 94°C, 15 seconds; 61°C, 30 seconds; 72°C, 2 minutes; 35 cycles. Genotyping primers were as follows: AML1-5sm 5'atagaaggaaacaaaaaggaaaggaggaa3'; AML1-3sm 5'tagcatatcacttccctgtagccgcactg3'; AML2-5sm 5'attgctctgtctctgatgatgttttgtcg3'; AML2-3sm 5'gcagcaatatttctaaagcatcgggttca3'; AML3-5sm 5'ctttagttccctctttcctctgctgtgat3'; AML3-3sm 5'ctgccaagaacgggaaaacaaacataaa3'; AML4-5sm 5'ttgcaagcggtagtccatataaatcctc3'; AML4-3sm 5'atgctaccgggagaacctaagtgaaatc3'; AML5-5sm 5'tctttagccacatcaatcattctcatcct3'; AML5-3sm 5'atcagcgtcaagttccattcctcctccac3'; JL-202 5'cattttataataacgctgcggacatctac3'; JL-270 5'tttctccatattgaccatcatactcattg3'; pROC-737 5'gggaattcactggccgtcgttttacaa3'. The wild-type loci were assayed with the above pairs. The insert was assayed using the following pairs: AML1-5sm or AML1-3sm with JL-270 or JL-202, AML2-5sm with pROC-737, AML3-5sm with JL-270 or JL-202, AML4-3sm with JL-270 or JL-202, AML5-5sm with JL-202. Insertions in the AML1 and AML5 loci were obtained from the University of Wisconsin alpha collection using their described protocol and are in the Wassiljewskia (Ws) ecotype background. The AML1 insertion was found in pool CSJ8-46-H35. The AML5 insertion was found in pool CSJ1091-H45. Insertions in the AML3 and AML4 loci were obtained from the University of Wisconsin Basta collection and are in the Ws background. The AML3 insertion was found in pool 67-6-F. The AML4 insertion was found in pool 18-2-H. The insertion in AML2 was obtained from the line 029713 from the Salk collection and is in the Columbia (Col-0) ecotype background. Aside from regions genetically linked to any of the insertion loci, the AML quintuple insertion allele homozygote lines are in the genomic background of a Col-0 × WS F 3 individual once backcrossed to the WS background. RNA extraction, RT-PCR RNA extraction was performed using TRIzol™ Reagent (Invitrogen) essentially according to manufacturer's instructions. Total RNA was treated with DNA- free ™ DNase (Ambion) and reverse transcribed using Omniscript reverse transcriptase (Qiagen) with an oligo-dT primer. Primer pairs spanning the insertion site for RT-PCR on all five lines are as follows: AML1 117–138 5'gtgatggatgcgattggataga3', AML1 556-534rc 5'attgtggcttcagctggtaactt3'; AML2 86–109 5'tttgcttctccgattctcttcctt3', AML2 456-435rc 5'agcatcgggttcaacatcttcc3'; AML3 548–568 5'gtagcggaggaggtcttgaat3', AML3 1060-1039rc 5'tctccttgatctcgccataaac3'; AML4 1932–1955 5'aagcggtagtccatataaatcctc3', AML4 2944-2927rc 5'tcccctgaatccgaccat3'; AML5 104–124 5'cgtgatcatcgtcggtgttgg3', AML5 1047-1024rc 5'ctcgacgaatttgtgatgcctctta3'. Reactions were performed using Takara ExTaq and 35 cycles of 94°C for 30 sec, 58°C annealing ( AML1,2,4 ) or 60°C annealing ( AML3,5 ) for 30 sec, and 72°C for one minute. Primer pairs amplifying a region 3' of the insertion site for RT-PCR on AML1 , AML2 , AML3 and AML5 insertion homozygotes are as follows: AML1 +1635 5'aggctctcgccgccctatta3', AML1 -2466 5'cgttgccaccttctcgctatt3'; AML2 +1779 5'accggggaacagtagtgaac3', AML2 -2107 5'ctgtcggcaagcatagaaag3'; AML3 +1756 5'tctggcctgctgctacaatgg3', AML3 -2326 5'cgccgacaagaagatgagaaaac3'; AML5 +1268 5'gcaacggcttccaacagtca3', AML5 -1869 5'acgaggcctaccattttcatacaa3'. All reactions were performed with a 59°C annealing temperature. Authors' contributions GHA conceived of the study and performed all experimental manipulations. MRH provided guidance and arranged funding for the project. GHA and MRH drafted the manuscript, and both authors approve the final manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC553980.xml
552335	Advances in wearable technology and applications in physical medicine and rehabilitation	The development of miniature sensors that can be unobtrusively attached to the body or can be part of clothing items, such as sensing elements embedded in the fabric of garments, have opened countless possibilities of monitoring patients in the field over extended periods of time. This is of particular relevance to the practice of physical medicine and rehabilitation. Wearable technology addresses a major question in the management of patients undergoing rehabilitation, i.e. have clinical interventions a significant impact on the real life of patients? Wearable technology allows clinicians to gather data where it matters the most to answer this question, i.e. the home and community settings. Direct observations concerning the impact of clinical interventions on mobility, level of independence, and quality of life can be performed by means of wearable systems. Researchers have focused on three main areas of work to develop tools of clinical interest: 1)the design and implementation of sensors that are minimally obtrusive and reliably record movement or physiological signals, 2)the development of systems that unobtrusively gather data from multiple wearable sensors and deliver this information to clinicians in the way that is most appropriate for each application, and 3)the design and implementation of algorithms to extract clinically relevant information from data recorded using wearable technology. Journal of NeuroEngineering and Rehabilitation has devoted a series of articles to this topic with the objective of offering a description of the state of the art in this research field and pointing to emerging applications that are relevant to the clinical practice in physical medicine and rehabilitation.	The potential impact of wearable technology on physical medicine and rehabilitation Understanding the impact of clinical interventions on the real life of individuals is an essential component of physical medicine and rehabilitation. While assessments performed in the clinical setting have value, it is difficult to perform thorough, costly evaluations of impairment and functional limitation within the time constraints and limited resources available in outpatient units of rehabilitation hospitals. Furthermore, it is often questioned whether assessments performed in the clinical setting are truly representative of how a given clinical intervention affects the real life of patients. While this observation has fostered a great deal of interest for the development and validation of outcome measures that largely rely on the use of questionnaires [ 1 ], researchers and clinicians have looked at recent advances in wearable technology intrigued by the possibility offered by this technology of gathering sensor data in the field [ 2 , 3 ]. Likely to be complementary to outcome measures, the use of wearable systems in the clinical management of individuals undergoing rehabilitation is very attractive because it provides the opportunity of recording quantitative data in the settings that matter the most, i.e. the home and the community. A number of clinical applications of wearable systems in physical medicine and rehabilitation emerged in the past few years. They range from simple monitoring of daily activities, for the purpose of assessing mobility and level of independence in individuals, to integrating miniature sensors to enhance the function of devices utilized by patients to perform motor tasks that they would be otherwise unable to accomplish. Monitoring functional motor activities was one of the first goals of research teams interested in clinical applications of wearable technology. The focus was initially on using accelerometers [ 4 - 8 ] or a combination of accelerometers and electromyographic sensors [ 9 ] to capture movement and muscle activity patterns associated with a given set of functional motor tasks. The set of tasks to be identified varied according to the clinical application. Their study was combined with monitoring systemic responses when the clinical assessment required combining motor activities and cardio-respiratory data such as in the clinical management of patients with chronic obstructive pulmonary disease [ 10 ]. A level of complexity was added when researchers started investigating motor disorders and the possibility of utilizing wearable technology to assess the effect of clinical interventions on the quality of movement observed while patients performed functional tasks. Two applications worth mentioning are the one to assess symptoms and motor complications in patients with Parkinson's disease [ 11 - 14 ] and the study of motor recovery in post-stroke individuals [ 15 - 17 ]. This shift from identifying functional motor activities to studying motor patterns associated with motor disorders generated significant interest for more complex ways to monitor movement, i.e. utilizing not only accelerometers but also gyroscopes and magnetometers or inclinometers. The combination of multiple sensors allows one to estimate the kinematics of movement [ 18 - 21 ] with a reliability that cannot be obtained by solely relying on accelerometers [ 22 ]. Finally, recent studies have been focused on integrating wearable, miniature sensor technology with orthoses, prostheses, and mobility assistive devices. Sensor technology is particularly appealing in these applications because it allows implementing closed-loop strategies that take advantage of the increased complexity and flexibility that robotics is contributing to the design of orthoses, prostheses, and mobility assistive devices. Namely, the characteristics of such devices can be constantly modified as a function of the task individuals are engaged into and environmental disturbances [ 23 , 24 ]. In all the emerging applications summarized above, either continuous recording of sensor data or at least monitoring over extended periods of time are necessary to design and implement an effective clinical intervention. Unobtrusive, wearable systems providing ease of data gathering and some processing capabilities are essential to achieve the objective of making the leap between the preliminary results obtained as part of the research carried on so far and the daily clinical practice of physical medicine and rehabilitation. Three areas of work are essential to achieve this objective: 1)the development of wearable sensors that unobtrusively and reliably record movement and other physiological data relevant to rehabilitation; 2)the design and implementation of systems that integrate multiple sensors, record data simultaneously from wearable sensors of different types, and relay sensor data to a remote location at the time and in the way that is most appropriate for the clinical application of interest; and 3)the development of methodologies to manipulate wearable sensor data to extract information in a clinically relevant manner to perform clinical assessments or control devices aimed at enhancing mobility in individuals with conditions that limit their level of independence. A series of papers have been assembled to provide the readership of Journal of NeuroEngineering and Rehabilitation with a description of the state of the art of the application of wearable technology in physical medicine and rehabilitation. Wearable sensors to measure movement and physiological signals A first set of the papers that have been assembled for publication on Journal of NeuroEngineering and Rehabilitation on the topic of wearable technology in physical medicine and rehabilitation has the objective of describing recent advances in wearable sensor technology. Two manuscripts describe attempts by different groups of measuring angular displacements for upper and lower extremity joints by embedding conductive fibers into the fabric of undergarments. The paper by Gibbs and Asada, entitled "Wearable conductive fiber sensors for multi-axis human joint angle measurements", reports encouraging preliminary results concerning monitoring lower limb joint displacements during ambulation by utilizing such technology. The manuscript by Tognetti et al, entitled "Wearable kinesthetic system for capturing and classifying upper limb gesture in post-stroke rehabilitation", describes the design and implementation of a system similar to the one proposed by Gibbs and Asada but geared toward monitoring movements of the upper extremities. The authors also explore the application of these wearable sensors to monitoring motor recovery in post-stroke individuals. Simone and Kamper focus their contribution on unobtrusively measuring finger movements in patients undergoing rehabilitation. Their manuscript "Design considerations for a wearable monitor to measure finger posture" summarizes the authors' recent work toward developing ways to record fine motor control tasks involving manipulation of objects requiring fine motor control of the hand and fingers. This technology has immediate application in patients such as post-stroke individuals undergoing rehabilitation that targets fine motor control skills. While initial research in the area of wearable technology was aimed at combining existing, miniature sensors with special fabrics or wireless technology, recent advances in this field have been focused on the development of sensing elements that can be even more easily embedded in clothing items. An example of such effort is reported in the paper by Dunne et al entitled "Initial development and testing of a novel foam-based pressure sensor for wearable sensing". This paper summarizes positive preliminary results by the research team aimed at measuring shoulder movements, neck movements, and scapular pressure. The sensing elements can also be used to monitor respiratory rate. Devoted to monitoring systemic responses is the last of the papers focused on wearable sensors. In this manuscript, Yan et al describe a new method to reliably measure heart rate and oxygen saturation. The paper is entitled "Reduction of motion artifacts in pulse oximetry by smoothed pseudo Wigner-Ville distribution" and demonstrates how advanced processing techniques may be necessary to derive reliable data when recordings are performed in the field. Wearable systems to gather data unobtrusively and reliably over extended periods of time A second area of research relevant to the application of wearable technology in physical medicine and rehabilitation concerns the integration of wearable sensors into systems. Following the seminal work by Park and Jayaraman [ 25 ], several researchers relied on conductive fabrics to deliver sensor data to a data-logger and then integrated it into a system that allowed remote access to the data. Other researchers explored the use of wireless technology as a means to relay wearable sensor data to a base station for data recording and remote access to clinically relevant information. Jovanov et al summarize recent advances by their research team toward developing body area networks in the manuscript entitled "A wireless body area network of intelligent motion sensors for computer assisted physical rehabilitation". Key points concerning the use of wireless technology in field monitoring of patients undergoing rehabilitation are the design of low-power transmission devices, the integration of multiple sensors, and the ability of providing processing capability that may reduce the amount of information to be transmitted. These issues are addressed in the above-referenced paper as well as in the manuscript by Sung et al entitled "Wearable feedback systems for rehabilitation". Sung et al describe a platform of wearable sensors recently developed by their team as well as potential applications currently under investigation. Clinical applications of wearable technology in physical medicine and rehabilitation A final set of papers is focused on applications that are relevant to physical medicine and rehabilitation. Sherrill et al describe in their paper entitled "A clustering technique to assess feasibility of motor activity identification in COPD patients via analysis of wearable-sensor data" a method to design classifiers of motor activities such as walking and stair climbing. The proposed technique relies on the examination of small datasets via clustering methods. Measures are derived from clusters associated with different motor activities to evaluate whether the set of wearable sensors and features derived from the recorded data are suitable to reliably identify the motor tasks of interest. Wang and Winters put the information gathered via wearable systems into a clinical context via processing that relies on neuro-fuzzy models. Their paper entitled "A dynamic neuro-fuzzy model providing bio-state estimation and prognosis prediction for wearable intelligent assistants" presents encouraging results indicating that the proposed method can put in the correct context dynamic changes observed in post-stroke individuals undergoing rehabilitation. Wang and Kiryu in their manuscript entitled "Personal customizing exercise with a wearable measurement and control unit" summarize their results on customizing machine-based exercise routines on the basis of physiological data that are continuously gathered from individuals performing such routines. Their results demonstrate the feasibility of a closed-loop system that optimally adapts workload. Dozza et al describe a wearable system designed to reduce body sway in individuals with severe vestibular problems. Their manuscript entitled "Influence of a portable audio-feedback device on structural properties of postural sway" summarizes positive results obtained with a prototype wearable system that utilizes audio-feedback to improve balance. Finally, Mavroidis et al describe how miniature sensor technology can be used to design a new generation of smart rehabilitation devices. Three devices are described in their paper entitled "Smart portable rehabilitation devices": a passive motion elbow device, a knee brace that provides variable resistance by controlling damping via the use of an electro-rheological fluid, and a portable knee device that combines electrical stimulation and biofeedback. These devices combine sensing technology and control strategies to enhance rehabilitation. Conclusion This collection of papers provides an up-to-date description of the state of the art in the field of wearable technology applied to physical medicine and rehabilitation. The field is rapidly advancing and numerous research groups have already demonstrated applications of great clinical relevance. The potential impact of this technology on the clinical practice of physical medicine and rehabilitation is remarkable. A significant shift in focus is possible thanks to wearable technology. While the main focus of clinical assessment techniques is currently on methods that are implemented in the clinical setting, wearable technology has the potential to redirect such focus on field recordings. This is expected to allow clinicians to eventually benefit from both data gathered in the home and the community settings during the performance of activities of daily living and data recorded in the clinical setting under controlled conditions. Complementarities are expected between field and clinical evaluations. Future research will surely address optimal ways to combine these two types of assessment to optimize the design of rehabilitation interventions.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC552335.xml
553995	Does angiotensin-1 converting enzyme genotype influence motor or cognitive development after pre-term birth?	Background Raised activity of the renin-angiotensin system (RAS) may both amplify inflammatory and free radical responses and decrease tissue metabolic efficiency and thus enhance cerebral injury in the preterm infant. The angiotensin-converting enzyme (ACE) DD genotype is associated with raised ACE and RAS activity as well as potentially adverse stimuli such as inflammation. The DD genotype has been associated with neurological impairments in the elderly, and thus may be also associated with poorer motor or cognitive development amongst children born preterm prematurely. Methods The association of DD genotype with developmental progress amongst 176 Caucasian children born at less than 33 weeks gestation (median birthweight 1475 g, range 645–2480 g; gestation 30 weeks, range 22–32; 108 male) was examined at 2 and 5 1/2 years of age. Measured neuro-cognitive outcomes were cranial ultrasound abnormalities, cerebral palsy, disability, Griffiths Developmental Quotient [DQ] at 2 yrs, and General Cognitive Ability [British Ability Scales-11] and motor performance [ABC Movement], both performed at 5 1/2 yrs. All outcomes were correlated with ACE genotype. Results The DD genotype was not associated with lower developmental quotients even after accounting for important social variables. Conclusion These data do not support either a role for ACE in the development of cognitive or motor function in surviving infants born preterm or inhibition of ACE as a neuroprotective therapy.	Background Delight over recent survival gains for the very premature infant has been tempered by the frequent presence of cerebral injury and developmental impairment. One quarter of those born before 26 weeks postmenstrual age (at least 11 weeks premature) show evidence of severe cerebral injury including cognitive dysfunction by 30 months of age [ 1 ]. Preterm children without any disability remain at risk of a range of motor, cognitive, behavioural and psychological deficits during childhood even if not born so close to the margin of viability [ 2 ]. To date, the pathophysiological processes leading to such impairment remain largely occult. In particular, cerebral imaging has failed to identify structural correlates of impaired higher function [ 3 ] although imaging can predict many cases of motor abnormality (such as cerebral palsy) due to the presence of periventricular white matter injury [ 4 ]. Three factors seem to play important roles in the aetiology of preterm cerebral injury. Firstly, exposure to inflammatory stimuli is associated with white matter injury and cerebral palsy in the preterm [ 5 ]. Secondly, reduced glucose and oxygen delivery to the developing brain (hypoxia-ischaemia: local cerebral or systemic) may cause excito-toxic neurotransmitter release followed by neuronal death [ 6 ]. Thirdly, free-radicals may damage the oligodendrocytes of white matter of the preterm brain [ 6 ]. Damage to the primitive white matter prevents the normal formation of grey matter connections which may influence cognitive development in childhood [ 7 ]. Candidate systems that might influence motor or cognitive outcome after premature birth are likely to be those which affect these responses. The human renin-angiotensin systems may be such a system. Angiotensin converting enzyme (ACE), a key component of the circulating (or endocrine) renin-angiotensin system (RAS), cleaves angiotensin I to yield the potent vasoconstrictor angiotensin II. In addition, ACE degrades vasodilator kinins. In these ways, endocrine RAS plays an important role in circulatory homeostasis. However, local RAS also exist in diverse human tissues including lung, myocardium, vasculature, lymphocyte and brain tissue. These are powerful regulators of mitochondrial respiration and whole-cell metabolism [ 8 ] and exert profound effects on whole-human metabolism and metabolic efficiency: elevated ACE may impair cellular aerobic metabolism [ 9 ]. RAS also plays a key role in the regulation of tissue inflammatory responses; ACE, through generation of angiotensin II, stimulates the synthesis of pro-inflammatory cytokines, including IL-6 which itself is thought to exert major neurocytotoxic effects with the genesis of functionally significant lesions in the developing preterm brain [ 5 ]. It has also been noted that the inhibition of RAS may reduce the effects of excitotoxic neurotransmitters and free radicals [ 10 ]. It is possible therefore that enhanced ACE activity may adversely influence the development of the child born prematurely. A common variant of the human ACE gene provides a tool to determine if ACE activity does influence developmental progress after preterm birth. The presence (insertion, or 'I' allele) rather than the absence (deletion, or 'D' allele) of a 284-base-pair fragment in the human ACE gene is associated with lower ACE activity in organs including both circulating inflammatory cells [ 11 ] and the circulation itself [ 12 ]. Given the likely causal association of pro-inflammatory responses, ischaemic-hypoxia, excitotoxic neurotransmitters, and free radical attack with impaired neuro-outcome; and given the potential role of increased RAS activity in amplifying these effects, we might expect the DD genotype (encoding raised ACE activity) to be associated with poorer neuro-developmental progress after pretem birth. Comparable findings have been described with respect to the deterioration of cognitive function in the elderly by some authors [ 13 - 15 ]. We have tested this hypothesis by studying the association of the ACE I/D polymorphism with measures of neuro-developmental progress at 2 and 5 1/2 years of age in children who had participated in a neuro-developmental outcome study (The Avon Premature Infant Project, APIP [ 16 ]). All the patients were born at less than 33 weeks postmenstrual age (normal gestation is 37–40 weeks). Methods Patients The study was approved by the ethical committees of Southmead Hospital and United Bristol Health Care Trust. Parental consent was obtained for participation in neurodevelopmental follow-up [ 16 ] (see below). Consent was not required for the genetic component of this study as all personal information was held separately from the genetic information and patients were identified only by study codes. All children were born at 32 weeks gestation or less, between December 1990 and July 1993 at Southmead Hospital or St. Michael's Hospital, Bristol. All had participated in the Avon Premature Infant Project (APIP) [ 16 ]. Briefly, this was a randomised controlled trial in which developmental support (Portage) or supportive counselling (parental adviser), each started at discharge and continued for up to 2 years, were found to confer some measurable (3–4 DQ points (below)) but clinically insignificant benefit to development at 2 years of age, when given in addition to appropriate primary care and community support, after adjusting for social variables. Neuro-developmental outcome The Griffiths Mental Development Scales, used to assess motor and cognitive performance, was performed at 2 years corrected age [ 17 ]. The Griffiths scales comprise five subscales, including personal and social, hearing and speech, locomotor, eye hand co-ordination and performance domains, from which is derived an overall developmental quotient (DQ). A lower Griffiths DQ reflects a poorer neuro-developmental performance, with a difference DQ of five points being clinically apparent. DQ was standardised originally to a mean of 100, with a standard deviation of 15, but secular drifts in population scores have resulted in a higher population mean. Thus for severe disability a score of 70 (-2 standard deviations (sd)) would indicate severe disability. Cognitive developmental progress at 5.5 years of age was assessed using the British Ability Scales [ 18 ]. The BAS-II was standardised in the early 1990s and was used to compute general cognitive ability (GCA) together with visuospatial, verbal and non-verbal subscales. The GCA is a developmental quotient, equivalent to an IQ estimate, normalised at 100 (sd +/- 15) in which a lower score again indicates poorer conceptual ability. The Movement ABC scales were used to assess manual dexterity, ball skills, and balance over ten tests at 5 1/2 years of age. Scores of each component are summed to produce a composite score ranging from 0–40, with high scores indicating a more impaired motor skills and 0 indicating normal skills. A psychologist performed the Griffiths Scales of Mental Development and a second psychologist performed the British Ability Scales (second edition) (BAS). The ABC Movement tests were performed by a trained research nurse. All assessments were blind to the child's neonatal course and subsequent progress. ACE genotyping DNA was extracted from the Guthrie card blood spots (newborn metabolic screening cards). ACE genotype was determined using 3-primer PCR amplification [ 9 ]. Primer ratios corresponded to 50 pmol of an I-specific oligonucleotide in a 20-υl reaction volume. The PCR was performed using Taq polymeraase yielding amplification products of 84 bp for the D allele, and 65 bp for the I allele. Amplification products were visualised using a 7.5% polyacrylamide gel stained with ethidium bromide. Genotyping was performed by staff blind to all clinical data. Study Size An estimate of sample size suggested that 144 patients would be needed for this study. The assumptions made for this calculation were that DD genotype infants had a mean DQ of 92.5 (1/2 SD below the norm) compared to a mean DQ of 100 in the ID+II group, assumed typical genotype distributions, and a significance of 0.05 with 80% power. Statistical analysis Data were stored in SPPS v9.0 for Windows. Lymphocyte [ 11 ] and tissue ACE [ 12 ] activity is primarily raised in DD genotype when compared to either ID or II genotype, and so data for those of DD genotype were compared to those from I-allele carriers. Categorical data were analysed by Chi square and continuous data by Student's T Test if normally distributed or Mann-Whitney U test as appropriate. Results Guthrie cards were located for 230 of 308 children. After exclusion of non-Caucasians and, at random, 1 child of any identical twin pairs (based on genotypes and gender) 176 babies with ACE genotype formed the study population (median birthweight 1475 g, range 645–2480 g; gestation 30 weeks, range 22–32) with follow-up data at 2 years. 122 of these also had follow-up at 5 1/2 years. The ACE genotype distribution was 49 [27.8%] DD, 73 [41.5%] ID, 54 [30.7%] II, demonstrated Hardy-Weinberg equilibrium, and was similar to that observed in the newborn term population from the same region of the UK (203 [24.1%] DD, 433 [51.5%] ID, 205 [24.4%) II). Baseline characteristics were independent of genotype, except that fewer individuals of DD genotype were from twin births ( p = 0.047) (table 1 ). There was no association between markers of neonatal cerebral injury: severe intraventricular haemorrrhage or white matter injury (table 1 ). There was no association with the presence of any disability at 2 years of age (DD 17% vs ID/II 15%, p = 0.65). Table 1 Perinatal and social factors DD Genotype (n = 49) ID/II Genotype (n = 127) No maternal antenatal corticosteroids 44 (80%) 112 (81%) No. of children from twin pregnancy* 4 (8%) 27 (21%) Male 32 (65%) 76 (60%) Gestation, weeks (± SEM) 29.7 (± 0.3) 30.0 (± 0.2) Birth weight, g (± SEM) 1453 (± 56) 1461 (± 34) Portage, parent adviser 17 (31%), 19 (35%) 46 (33%), 42 (30%) Severe intraventricular haemorrhage 5 (11%) 7 (6%) White matter injury 7 (14%) 14 (11%) Maternal age (± SEM) 27.2 (± 0.8) 27.4 (± 0.8) Manual occupation 28 (57%) 76 (60%) Maternal car use 30 (61%) 75 (60%) Mother educated beyond 16 yrs. 17 (35%) 48 (38%) *p = 0.047 (Fisher's Exact Probability Test) Continuous data is shown as mean (± standard error of mean). Measures of developmental cognitive and motor outcome were entirely independent of genotype (table 2 ). The findings were unchanged after post hoc subgroup analysis of singletons, infants with normal cranial scans, amongst children without disability and after adjusting for potential influential variables (including twin birth) using multiple regression (data not shown). Table 2 ACE genotype and developmental performance at 2 and 5 1/2 years of age. Data shown is mean (± SEM). Developmental tests DQ for DD DQ for ID/II p Griffith DQ at 2 years 96.2 (3.1) 96.3 (1.3) 0.95 Locomotor subscale 92.7 (2.7) 92.4 (1.3) 0.92 Personal & social subscale 101.9 (3.0) 101.0 (1.6) 0.80 Hearing and speech subscale 92.9 (4.1) 94.0 (2.1) 0.80 Eye hand co-ordination subscale 90.8 (3.1) 92.8 (1.2) 0.46 Performance subscale 102.2 (4.3) 101.3 (1.6) 0.79 Griffith DQ at 2 years (adjusted for social variables) 100.0 (0.9) 99.3 (0.6) 0.43 ABC Movement summative score 8.1 (1.8) 8.0 (0.9) 0.97 GCA at 5 1/2 years 99.2 (3.4) 100.2 (2.0) 0.80 Verbal ability subscale 98.0 (4.0) 103.2 (1.7) 0.22 Pictoral ability subscale 99.9 (3.3) 98.7 (1.7) 0.99 Spatial ability subscale 98.4 (3.3) 97.3 (1.9) 0.67 Discussion After a search of Embase and Medline we believe that this study is the first to attempt to dissect out the contribution of genetic variation in the ACE gene to developmental progress after pre-term delivery. Despite much physiological and biochemical evidence to support our hypothesis, we found that ACE DD genotype was not associated with adverse long term developmental outcome in infants of < 33 weeks gestation in this study. These data are perhaps at variance with previous studies of Alzheimer's disease, age-associated memory impairment and vascular dementia, all of which have implicated the ACE D allele in having a role in mental decline [ 13 - 15 ]. However this is not a universal finding. Furthermore although ACE inhibitors appear to reduce inflammatory responses, ischaemic effects, and excitotoxic and free radical induced injury [ 10 ], angiotensin II does not (indeed angiotensin II may actually enhance ischaemic and excitotoxic neural injury via the AT2 receptor). In addition, both captopril and losartan (RAS inhibitors) appear to improve cognitive performance in mice [ 19 ] and humans [ 20 ]. It should be noted however that little is known about the ontogeny of the RAS in the human foetus. Certainly RAS (and angiotensin II receptors in particular) play a role in blood-brain barrier and central nervous system development in mice, and alterations in RAS receptor expression over foetal and neonatal life are recognised. It is thus possible that developmentally regulated patterns of AT1 receptor expression might offer some level of protection against the potentially detrimental effects of ACE-mediated angiotensin II synthesis. Although there may be similar molecular pathways that effect cerebral injury in the preterm infant and the elderly, ontological differences in the expression of genes involved in predisposition to neural injury are well described. In particular reactive production of nitric oxide may be enhanced in the elderly and the ability to protect the brain from oxidants may be reduced in the elderly (22). Thus the effect of any one polymorphism, with a relatively minor effect, may be swamped in the newborn infant by other protective mechanisms. The lack of any association between ACE genotype and scores of developmental progress was also surprising because we have demonstrated an association between DD genotype and markers of poor cardio-respiratory instability in the perinatal period in this patient group [ 21 ]. This association (between genotype and worse early cardio-respiratory status) could predispose to death, which would in turn weaken any association (if it exists) between DD genotype and worse developmental quotients. It is of course possible that our sample size was insufficient to demonstrate any association with ACE genotype and developmental progress. However, similar-sized studies have been sufficient to demonstrate an association between ACE D allele and cognitive decline in the elderly [ 13 - 15 ], and power calculations suggested we had enough patients to demonstrate at least a trend. If an undetected genotype-association does exist such an effect is weak. Conclusion We cannot support an association of ACE genotype with cognitive or motor development in survivors born preterm or, thus, the use of RAS inhibition as a neuroprotective agent in the preterm. Given the current lack of understanding of the mechanisms leading to cerebral injury and subsequent impairment – particularly of higher function – in such patients, further genetic association studies of other candidate genes are warranted. List of abbreviations used ACE, angiotensin-1 converting enzyme; DQ developmental quotient, BAS, British ability scales (second edition); GCA, general cognitive ability; RAS, renin angiotensin system; PCR, polymerase chain rection. Competing interests The author(s) declare that they have no competing interests. Authors' contributions DH, HM, AW, NM conceived the study and its design and wrote the manuscript. DH, DD and SD performed data collection, DNA extraction and PCR and participated in analysis of the data with SH and HM. NM reviewed all cranial imaging. All authors participated in the writing of the manuscript and approved the final manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC553995.xml
544543	Unexpressed but Indispensable—The DNA Sequences That Control Development	null	Amidst the hoopla over the exact number of genes we have in our genome—more than a fruitfly, fewer than a rice plant—a more fundamental genetic truth has often been obscured. The expression of 20,000–30,000 genes is under the control of an uncounted host of non-coding sequences, which bind transcription factors and thereby regulate when and where genes are expressed. Unlike coding sequences, whose signatures are easy to spot, the characteristic features of non-coding regulatory elements are largely unknown, making their discovery by simple sequence analysis difficult. In this issue, Greg Elgar and colleagues attack this problem by comparing the non-coding sequences of the human and the pufferfish. Since the last common ancestor of these two species existed 450 million years ago, the authors reasoned that non-coding sequences conserved between them are likely to be fundamental to vertebrate development. Through sequence alignment with increasingly strict criteria, they identified 1,373 highly conserved non-coding elements (CNEs), with an average length of about 200 base pairs. The average sequence match is 84%: not perfect, but much higher than for coding regions shared by humans and pufferfish. A quick check showed that virtually all the sequences also occurred in rodents, chickens, and zebrafish, but not in the nematode, fruitfly, or even the sea squirt, a primitive non-vertebrate chordate. Highly conserved vertebrate non-coding elements direct tissue-specific reporter gene expression CNEs are not spread uniformly throughout the genome. Instead, they are bunched together in fewer than 200 clusters, most of them in close proximity to genes implicated in transcriptional regulation or development. This clustering of CNEs suggests they may not only attract transcription factors, but may also influence the local topology of the DNA, thereby controlling access to their associated gene. Several clusters also appear in regions without any known genes—the identification of these clusters might lead to the discovery of new developmentally significant genes. While “in silico” discoveries such as this can be the jumping-off point for whole new areas of investigation, their validity must be tested “in aqua,” in the wet biology of real organisms. For this Elgar and colleagues chose the zebrafish, because its transparent embryo is ideal for observing developmental events. They injected individual CNEs into embryos, along with a green fluorescent protein (GFP) reporter. By day two of development, 23 out of 25 CNEs injected had upregulated GFP expression, indicating interaction of these sequences with endogenous transcription factors. Different CNEs caused different regional patterns of expression, in keeping with their presumed roles in distinct developmental processes. The discovery of these developmentally important sequences opens several avenues of new research. For example, analyzing the sequence and location of these CNEs may help point the way to other non-coding elements that remain undiscovered. It is also likely that mutations in these critical sequences cause human diseases. Studying how such mutations drive development astray may lead to better understanding not only of these diseases, which are likely to be rare, but also of normal human development.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC544543.xml
544542	Neuronal Encoding of Texture in the Whisker Sensory Pathway	A major challenge of sensory systems neuroscience is to quantify brain activity underlying perceptual experiences and to explain this activity as the outcome of elemental neuronal response properties. Rats make extremely fine discriminations of texture by “whisking” their vibrissae across an object's surface, yet the neuronal coding underlying texture sensations remains unknown. Measuring whisker vibrations during active whisking across surfaces, we found that each texture results in a unique “kinetic signature” defined by the temporal profile of whisker velocity. We presented these texture-induced vibrations as stimuli while recording responses of first-order sensory neurons and neurons in the whisker area of cerebral cortex. Each texture is encoded by a distinctive, temporally precise firing pattern. To look for the neuronal coding properties that give rise to texture-specific firing patterns, we delivered horizontal and vertical whisker movements that varied randomly in time (“white noise”) and found that the response probabilities of first-order neurons and cortical neurons vary systematically according to whisker speed and direction. We applied the velocity-tuned spike probabilities derived from white noise to the sequence of velocity features in the texture to construct a simulated texture response. The close match between the simulated and real responses indicates that texture coding originates in the selectivity of neurons to elemental kinetic events.	Introduction One goal of sensory systems neuroscience is to understand how the representations of complex, natural stimuli arise from the basic response properties of neurons. The present experiments explore the representation of textures in the rat somatosensory system. Rats have texture discrimination capacities rivaling those of humans [ 1 ]. In rats, as in humans [ 2 ], object exploration in the tactile modality derives from active palpation. Thus, rats create sensory signals by sweeping their whiskers across surfaces in a rhythmic forward-backward cycle with a frequency ranging from 5 to 15 Hz [ 1 , 3 , 4 , 5 ]. Several hundred primary afferent fibers—“first-order neurons”—innervate specialized receptors on each whisker shaft [ 6 ], and these are excited by whisker movement. Signals travel along the sensory nerve, past the cell body in the trigeminal ganglion, to the brain stem. Here the first synapse is located. The axons of second-order neurons cross the brain midline and travel to the thalamic somatosensory nuclei, where the second synapse is located. Thalamic neurons project to the primary somatosensory cortex, conveying information to layer IV cell populations called “barrels” [ 7 , 8 ]. There have been no reports concerning the cortical or subcortical neuronal activity generated by whisking along irregular surfaces, and the differences in activity associated with two surfaces remain unknown [ 9 ]. However, recent work suggests the framework for a texture coding model. For nonnatural whisker deflections such as ramps [ 10 ], sinusoids [ 11 , 12 ], and temporally unstructured movement [ 13 ], first-order sensory neurons and cortical neurons emit spikes with probabilities that increase in proportion to stimulus velocity. This raises the possibility that neurons represent texture by encoding the kinetics of whisker vibrations. However, key elements of the model are untested. Does whisker movement across different textures produce distinct vibrations? If so, do neurons in the central pathway reliably report these vibrations? Through what coding mechanisms? To answer these questions, the model must be challenged under conditions where the sensory input is precisely controlled and yet resembles what occurs during natural tactile behavior. Guided by this strategy, in anesthetized rats we produced whisker movements across textures while measuring vibrations of the whisker shaft. We then played back the identical vibrations to other rats, and measured the neuronal activity at two stages of the sensory pathway—the first-order neurons that innervate the sensory receptors, and the barrel cortex neurons, which are the first site of cortical integration. Texture discrimination depends on the integrity of the cortical barrels [ 14 ]. By measuring the activity of trigeminal ganglion neurons (the cell bodies of the first-order neurons), we investigated how the sweeping motion of whiskers along a surface is converted to a neuronal impulse code. By measuring activity in the cortex, we explored the neuronal representation that rats rely on to judge the identity of external objects [ 15 , 16 ]. Comparing two levels of the pathway, we reveal the transformation of neuronal signals at successive levels of integration, and show how the neuronal signals emerge from elemental feature extraction. Results Kinetic Signatures of Textures The experimental strategy ( Figure 1 ) was to collect records of the natural movement of whiskers across surfaces ( Figure 1 A) and use them as a stimulus set to probe the neuronal representation of texture ( Figure 1 C). In one group of rats ( n = 3), we electrically stimulated cranial nerve VII, generating 8-Hz whisking movements [ 17 , 18 ] that resemble whisker trajectories in awake rats [ 4 ]. Meanwhile, whisker displacements transmitted to the receptors in the follicle were measured by an optical sensor placed 1 mm from the skin. The vertical and horizontal channels of the sensor ( Video S1 ) reported whisker position with less than 3-μm spatial and 0.13-ms temporal resolution. Movements were measured under different conditions ( Figure 1 B, “texture” column): whisking with no object contact (“free whisk”), whisking on compact disk surface (smooth), and whisking on sandpapers of four different grades: P1200, P400, P280, P100 (from fine-grained to coarse-grained; Table 1 ). The surface was oriented so that the whisker rested on it and remained in contact during the entire whisk trajectory. The proximal edge of the surface was 7 mm from the base of the whisker. The illustrated data ( Figure 1 ) come from whisker C3 in rat EW3. Figure 1 Collection and Playback of Texture Library (A) Whisker vibration data were collected during “electrical whisking,” induced by stimulation of the facial nerve (1) with pulse trains (2) in rat EW3. An optical sensor, shown schematically by two orthogonal light paths (3), monitored vertical and horizontal whisker motion of whisker C3. (B) “Texture” column: Photographs of the 5 surfaces used. “Trajectory” column: Sample whisker trajectories (first whisk of trial 50) associated with free whisking and the five surfaces. Each point, separated by 1 ms, gives the horizontal and vertical position; the trajectory begins with protraction (P) at t = 0 and terminates 125 ms later at the end of retraction (R). Speed is given by the color of each point. Note the irregularities—jumps, stops, and starts—induced by whisking on sandpaper. “Velocity profile” column: Whisker trajectories displayed according to the horizontal and vertical velocities (V H and V V, respectively). P refers to protraction phase (positive V H ), and R to retraction phase (negative V H ). In this and all figures, V H and V V were calculated 7,634 times per second. “Velocity spectrogram” column: Velocity spectrograms for each texture (see Materials and Methods ). (C) Playback of the whisker trajectories to a second group of rats through a piezoelectric motor (4), shown schematically by the horizontal and vertical arrows at the base of the whisker. Table 1 Parameters of the Sandpapers See http://www.fepa-abrasives.org ND, no data Under free whisking conditions the trajectory was a smooth ellipsoid [ 4 ], the principal axis aligned with protraction and retraction movements (P and R, respectively, in Figure 1 B, “trajectory” column). As the rat whisked on the compact disk surface, the trajectory was similar but covered a more restricted vertical range. In contrast, whisking across grainy surfaces produced irregularities in the trajectory, and each texture was associated with a characteristic whisker shaft vibration. These distinct “kinetic signatures” are evident in the velocity profile—that is, the temporal sequence of velocity features across the course of a whisk ( Figure 1 B, “velocity profile” column). Each velocity profile covers 125 ms (one complete forward and backward whisk) and consists of two histograms—horizontal (V H ) and vertical (V V ) velocity. For V H , whisker protraction (forward movement) is positive and whisker retraction (backward movement) negative. For V V , upward movement is positive and downward movement negative. To better visualize the time-varying frequency content of the velocity profiles, the velocity spectrograms are also plotted ( Figure 1 B, “velocity spectrogram” column). The spectrograms were formed by computing the magnitude of all sinusoidal components (0–500 Hz) of the velocity profile in a 6-ms wide window, and then sliding the window with 0.13-ms time steps (see Materials and Methods ). The velocity profile and the velocity spectrogram, taken together, illustrate the kinetic features that make each texture unique—the duration and frequency content of each velocity peak, as well as the number of peaks and the temporal spacing between them. In a different group of rats, the texture-induced vibrations were played back to the base of a whisker ( Figure 1 C and Video S2 ) and neuronal activity was recorded. Construction of the stimulus sequence off-line allowed us to smoothly “stitch together” the vibrations: The transitions between textures and free whisks were always inserted at time zero, corresponding to the point of maximal retraction, when whisker velocity was zero. The replay was an accurate replica of the motion recorded during electrical whisking ( Figure S1 ). Receptor and Cortical Coding Properties The physiological dataset (seven rats) consists of six first-order neuron recordings, five cortical cluster recordings, and seven “paired” recordings—simultaneous first-order neuron and cortical cluster. The principal result is that time-varying neuronal activity in the trigeminal ganglion and cortex captured the kinetic features of the texture-induced vibrations. From the same texture library given in Figure 1 (rat EW3, whisker C3), Figure 2 A gives the velocity profile, averaged across 100 trials, for two free whisks (−250 to 0 ms) followed by two whisks on P280 sandpaper (0 to 250 ms). Note the distinct kinetic signatures of whisker movement: Unlike free whisking, the coarse (P280) sandpaper caused irregular bursts of high and low velocity, particularly during whisker retraction. The response of one first-order neuron (named Zurvan) is shown in Figure 2 B as a raster plot of 100 trials and in Figure 2 C as a peristimulus time histogram (PSTH). Several coding properties are evident: (i) The first-order neuron fired a greater number of spikes for the coarse texture than for free whisks; (ii) spikes were closely aligned to instants in which the whisker moved at high velocity (blue arrowheads); (iii) it fired in a reproducible manner across trials—the spikes were aligned; and (iv) it was selective to whisker retraction (did not fire for high-velocity protractions; red arrowhead). Figure 2 Sensory Receptor and Cortical Coding Properties (A) V H and V V for two free whisks followed by two P280 whisks. The labeling conventions are as in Figure 1 B. Each presented trial was unique due to small variations in whisker trajectory even on the same surface ( Figure 7 ); the illustrated velocity profiles are the averages of 100 trials. The red arrowhead indicates the time of the first V H peak during whisker protraction on P280; blue arrowheads indicate the times of the three V H peaks during whisker retraction on P280. (B) Raster plot of first-order neuron aligned with the whisker trajectories, in response to 100 unique trials. Stimuli were applied to whisker E4. (C) PSTH of first-order neuron with 0.2-ms bins. Blue arrowheads indicate the times of maximum response to the three peaks in retraction velocity. The red arrowhead indicates the expected time of response to the peak in protraction velocity; however, the neuron did not respond to whisker protraction. (D) Raster plot for the cortical neuron cluster recorded simultaneously with the first-order neuron. (E) Two cortical PSTHs, both with 2-ms bins. The upper PSTH corresponds to the raster plot in (D); the lower PSTH is from a second cortical neuron cluster recorded simultaneously at a neighboring electrode (distance 560 μm). Blue and red arrowheads indicate the times of maximum response to the peaks in whisker protraction and retraction velocity, carried down from (A). The cortical neuron clusters responded to high velocities for both protraction and retraction. Because the first two peaks in retraction velocity were separated by just 7 ms, the resulting peaks in cortical response were fused. All PSTHs are extended to 260 ms to show responses to the final velocity feature. Two barrel cortex neuron clusters were recorded simultaneously with the first-order neuron, allowing direct comparison of different stations along the sensory pathway. Figure 2 D shows the raster plot for one of the cortical clusters, while Figure 2 E shows PSTHs for both cortical clusters. Like the first-order neuron, the cortical clusters responded to high velocities (arrowheads, Figure 2 A, 2 C, and 2E) and, as a result, fired a greater number of spikes for P280 sandpaper than for free whisks. Key differences from the first-order neuron are clear: (i) The cortical neuron clusters fired in a less reproducible manner across trials—there was more variability in the number of spikes per whisk and in the temporal alignment of spikes; and (ii) they fired for both whisker protraction and retraction (red and blue arrowheads, respectively). The selectivity of first-order neurons and cortical clusters for the direction of whisker movement is described in more detail in Figure 3 . Figure 3 Directional Selectivity in First-Order and Cortical Neurons (A) Mean spike count per whisk for ten first-order neurons separated into protraction and retraction phases. Responses to free-whisk and all textures were combined, giving a total of 8,000 whisks. First-order neurons are arranged from left to right according to their retraction:protraction spike count ratio. Five first-order neurons preferred retraction, three preferred retraction, and two responded to both phases. Principal whisker of each neuron is indicated. The neuron Zurvan is indicated by an asterisk. (B) Same analysis for 12 cortical clusters. Individual cortical neuron clusters did not present a clear preference for either retraction or protraction. Conclusions about single unit directional selectivity cannot be drawn, however, because the directional selectivity of any cluster must always be less than that of the most selective single unit in the cluster. The neuron cluster ( Figure 2D, 2E ) recorded simultaneously with Zurvan is indicated by an asterisk. Texture Coding by Firing Rate To permit sensory discriminations, some properties of neuronal firing must vary systematically from texture to texture. Earlier work [ 11 , 12 ] showed that neuronal firing rate, in response to sinusoidal whisker movement, is dictated by mean vibration speed, proportional to the product of amplitude and frequency, Xω (referred to in previous publications as Af ). The generalization of Xω to the natural, texture-induced vibration is 〈\| X(ω,τ) \| ω 〉 Ω,Τ , a quantity known as “equivalent noise level” (see Materials and Methods ). In Figure 4 , we compare response magnitude for the full population of first-order and cortical neurons to each texture's equivalent noise level. The Pearson correlation coefficient between equivalent noise level and spike count was 0.93 for the first-order neurons and 0.99 for the cortex. This finding indicates that neuronal spike count is a function of the magnitude of the composite frequency components of the whisker vibration, whether the stimulus is a simple sinusoid (where all the power is at a single frequency) or a complex, texture-induced vibration. Figure 4 Texture Coding by Firing Rate Equivalent noise level (plus SD) of texture-induced vibrations averaged across 100 trials of 500 ms each (see Materials and Methods ). Average spike count per trial (plus SD) pooled from ten first-order neurons and 12 cortical neuron clusters. Note separate scales for spike counts of neurons recorded in Ganglion and Cortex. Texture Coding by Firing Patterns Since different sandpapers can induce vibrations with similar equivalent noise levels and thereby evoke similar spike counts (e.g., P400, P280, and P100 in Figure 4 ), we must expect additional texture coding mechanisms to be at work. We therefore examined more closely the neurons shown in Figure 2 , on the hypothesis that spike patterns might carry texture-specific information. V H and V V profiles associated with two whisks on each texture are plotted ( Figure 5 A), together with PSTHs for the first-order neuron ( Figure 5 B) and the cortical neuron cluster ( Figure 5 C). Textures evoking similar spike counts due to similar equivalent noise levels were readily distinguished by spike patterns. The patterns arose from the alignment of spikes to the velocity profile of the input vibration. An assessment of spike alignment to other stimulus features (whisker position and whisker acceleration) is given in Figure 6 and indicates that these features were reported less reliably than whisker velocity. The first-order neuron reported the velocity profile for whisker retraction, while the cortical neuron cluster reported both protraction and retraction profiles, albeit with lower fidelity to individual velocity features. Figure 5 Texture Coding by Firing Patterns (A) V H and V V for two whisks on texture P400 (left), P280 (middle), and P100 (right). Each illustrated velocity profile is the average of 100 unique profiles. (B) First-order neuron PSTHs (0.2-ms bins) aligned with the whisker trajectories. (C) Cortical PSTHs (2-ms bins). PSTHs are extended to 260 ms. The arrowheads on the left side of PSTHs indicate mean firing rates. Figure 6 Test for First-Order Neuron Encoding of Position and Acceleration To investigate whether first-order neurons represented stimulus features other than velocity, we repeated the same analysis as in Figure 5 , in relation to whisker position (A) and acceleration (B), because it has been suggested that neuronal activity is determined by these stimulus parameters [ 10 , 13 ]. Alignment between the PSTH (C) and stimulus position or acceleration revealed no consistent correlation. For texture P100, the boxes extending across A, B, and C highlight the absence of correlation. For example, two periods with similar positions produced first no spikes (red-outlined box on left) and then a large response (red-outlined box on right). Moreover, high acceleration (left box) produced no spikes, while lower levels of acceleration (right box) produced a large response. For this neuron, only velocity was encoded. Sources of Neuronal Variability In a number of sensory modalities, first-order neuron responses can be remarkably reliable when a stimulus is presented repeatedly [ 19 ], whereas cortical responses vary across trials [ 20 ]. It is of interest to elucidate the mechanisms that permit reliable first-order neuron responses and, by the same token, to identify the sources of trial-to-trial variability among cortical neurons. In the data shown so far, the 100 trials for a given texture were composed of 400 unique whisks (four whisks per trial). Each whisk differed in the minute details of its trajectory ( Figure 7 ). To discover the origin of neuronal variability, we selected trial 50 for each texture and repeated the four-whisk sequence 100 times. If neuronal variability originates purely in stimulus variability, it will disappear across repeated trials; variability due to internal brain fluctuations, however, will remain. Figure 7 Velocity Profile Variability across Trials Ten successive trials are shown (numbers 46–55), each trial composed of the final two free-whisks (–250 to 0 ms) and the first two whisks on P280 (0 to 250 ms). Free whisk velocity profiles varied little across trials. When the whisker swept across P280 repeatedly, the fundamental kinetic signature was conserved (e.g., the three peaks in retraction velocity for P280) but minute details of the profile varied—note, for example, the velocity event (red asterisk) that occurred uniquely on trial 50. The velocity profile for the final two free whisks (−250 to 0 ms) and the first two P280 whisks (0 to 250 ms) of trial 50 is given in Figure 8 A. The response of the first-order neuron is shown as a raster plot ( Figure 8 B) and a PSTH ( Figure 8 C). These can be compared to responses of the same cell in Figure 2 B and 2 C. Response was nearly identical on each trial, because of the precise temporal alignment of spikes on the high-velocity events. Indeed, some stimulus features evoked 0.7–0.8 spikes per bin per trial, meaning that neuronal jitter fell within the 0.2-ms PSTH bin size. Figure 8 Sources of Neuronal Variability (A) V H and V V across the final two free whisks and the first two P280 whisks of trial number 50. Here, as in Figure 2 , the red arrowhead indicates peak whisker velocity during protraction, and blue arrowheads indicate the peak whisker velocities during retraction. (B and C) First-order neuron raster plot (B) and PSTH (C), aligned with the whisker trajectories, for 100 stimulus repetitions. Due to the temporal precision of neuronal responses, the vertical scale of the PSTH has been altered (compare to Figures 2 and 5 ) to reflect the large numbers of spikes within single bins. (D) Cortical neuron cluster raster plot. (E) Two cortical PSTHs from activity recorded simultaneously with the first-order neuron. The upper PSTH corresponds to the raster plot in (D); the lower PSTH is derived from a second cortical neuron cluster recorded simultaneously at a neighboring electrode (distance of 560 μm). PSTHs have 0.2-ms bins for the first-order neuron and 2-ms bins for the cortical neuron clusters. All PSTHs are extended to 260 ms to show responses to the final velocity feature. Response peaks are signaled by red and blue arrowheads according to the velocity events that evoked them. In Figure 8 D and 8 E, the cortical response to repeated trials is presented. Direct, quantitative comparisons between the variability of first-order responses and that of cortical responses cannot be made, because the cortical recordings were made from multi-neuron clusters. However, the cortical response to repeated trials can be compared to the same cluster's response to 100 unique trials in Figure 2 D and 2 E. Eliminating trial-to-trial variability in the timing of stimulus features reduced but did not eliminate neuronal jitter. The remarkable response locking of first-order neurons to stimulus features is further highlighted in Figure 9 . Unlike Zurvan, the illustrated neuron was selective to whisker protraction. For texture P280, the velocity histograms ( Figure 9 A) contained well-separated peaks during protraction. With 100 repetitions of trial number 50, the raster plot ( Figure 9 B) and PSTH ( Figure 9 C) yielded discrete response peaks aligned with high-velocity protraction events. Applying the green horizontal line to the PSTH as a threshold, we extracted six separable response clusters. Every cluster contained exactly 100 spikes resulting from one spike per trial for each velocity event. The final response was evoked by a clear protraction event (red asterisk in Figure 9 A) and was selected for closer inspection (red inset). Here, the 100 spikes spanned a range of 0.38 ms; standard deviation (SD) in spike time was 0.08 ms. For the five preceding response clusters, spike time SDs were 0.13, 0.09, 0.09, 0.13, and 0.11 ms. The minimum measured spike time SD in the dataset was 0.07 ms. These must be taken as underestimates of spike time precision, given that they include measurement noise inherent to the recording system (e.g., the SD in spike time caused by digitizing the action potential threshold crossing time at 30 samples per ms is nearly 0.02 ms). Figure 9 Precision of a First-Order Neuron (A) V H and V V across the first two P280 whisks of trial number 50 (see Figure 7 ). (B and C) Raster plot (B) and PSTH (C) of the first-order neuron for 100 repetitions of the stimulus given in (A). Inset in red frame shows a magnified view of spikes emitted in response to a single velocity event (red asterisk in [A]) and their SD in time. The same measurement of jitter was carried out for each of the response peaks that surpassed the green horizontal line (see text). From these observations we conclude that, under our experimental conditions, the trial-to-trial response variability of first-order neurons is caused exclusively by stimulus jitter, whereas that of cortical neurons results mainly from variations across time in sensory integration, and must emerge at some integration site between the trigeminal ganglion and cortex. A question of current interest is whether the variability in cortical responses results from noise and imprecision in neuronal integration [ 20 ], or else reflects functionally significant modulations in responsiveness [ 21 , 22 , 23 ]. From Response Properties to Natural Responses We hypothesize that the firing patterns of first-order and cortical neurons during presentation of textures can be explained by their extraction of elemental features from the complex input signal, and that these elemental features are bursts of high velocity. To test this directly, we presented a “white noise” stimulus in which the two stimulus features V H and V V varied randomly across time. Responses to noise stimuli allowed us to quantify velocity sensitivity and then to generate simulated spike trains based on the sequence of velocity events in the actual texture-induced vibrations. Finally, comparison between simulated and observed responses reveals the extent to which the responses to natural stimuli are explained by neuronal selectivity to velocity features: If simulated responses closely resemble real responses, we can conclude that neurons are in fact operating on natural texture stimuli according to their tuning to elemental kinetic events. Typically, neuronal tuning curves are mapped out using an “unbiased” stimulus—a stimulus that avoids the temporal correlations present in natural stimuli. The first step, therefore, was to map out how first-order and cortical neurons encode whisker kinetic features when these features are extracted from the context of the natural stimulus. We applied a stimulus that varied randomly in velocity—Gaussian velocity noise—and therefore was not constrained by the velocity patterns present in texture trajectories (see Materials and Methods ). Figure 10 A tracks V H and V V (white circles) across 5 ms of filtered white noise. Figure 10 Velocity Tuning Curves and Simulated Texture Responses (A) A 5-ms trajectory of velocity white noise. Radial coordinates give V H , V V . Velocity space was subdivided such that each segment included the same number of events (3,435,300). One segment (red outline) is selected for further explanation (see text). (B) 100-ms ganglion and cortical spike train aligned below occurrences of the velocity event of interest (red bar). After each such event, spike times were accumulated to build up a spike probability profile. (C) First-order neuron spike probabilities, given by color scale, in relation to joint A,R events. To estimate the tuning curve in finer detail, the number of angles was increased to 20. Each segment now contains about 1,374,120 velocity events. One P280 whisk trajectory is superimposed. (D) Spike probabilities for cortical neuron cluster, given by color scale, in relation to joint A,R events. One P280 whisk trajectory is superimposed. (E) Simulated raster plot for first-order neuron and simulated (black) and real (red) PSTHs. (F) Simulated raster plot for cortical neuron cluster and simulated (black) and real (red) PSTHs. We then constructed firing probability profiles in relation to millions of occurrences of each velocity event, such as the velocity event in the fifth angular sector and ninth radial sector, or A 5 ,R 9 (red outline in Figure 10 A). Figure 10 B shows how response probabilities were constructed in relation to this particular velocity event. The first trace shows the occurrences of A 5 ,R 9 across a 100-ms window—the first occurrence of the event (asterisk in Figure 10 A) corresponds to the crossing of A 5 ,R 9 in Figure 10 A (also marked by an asterisk). Below, first-order sensory neuron and cortical spike times are shown. After 10 min of stimulus noise, the postevent spike probability profiles associated with A 5 ,R 9 and all other events could be constructed. For the first-order neuron Zurvan, spike probabilities in the 1–2-ms postevent interval are given in Figure 10 C for all joint events ( A,R ). To construct the neuron's “tuning curve” in finer detail, velocity space was subdivided into 20 angular and 10 radial segments. The neuron emitted spikes with increasing probability as velocity increased, but only for restricted directions, preferring high speeds that combined retraction (negative horizontal velocity) and upward movement (positive vertical velocity). For the simultaneously recorded cortical neuron cluster, spike probabilities in the 5–20-ms poststimulus interval are given in Figure 10 D; like the first-order neuron, the cortical cluster emitted spikes with increasing probability as speed increased, but its directional selectivity was less pronounced and was radially symmetric. Can the neurons' responses to complex, natural stimuli be explained as the outcome of these elemental tuning properties? To find out, we projected whisk velocity trajectories upon the white noise-derived tuning curves. One whisk (first whisk of trial 50) on P280 sandpaper is depicted on both tuning curves. The first observation is that the intersection of the velocity trajectory with the tuning curves explains why the first-order neuron was selective for whisker retraction while the cortical cluster was directionally nonselective (see Figures 2 and 5 ). Figures 10 E and 10 F show the simulated responses of the first-order neuron and cortical cluster to two P280 whisks, delivered 100 times. The simulation was of 100 unique trials (see Figure 2 ) rather than repeated trials (see Figure 8 ). Thus, on each trial the minute details of the whisks gave rise to a unique sequence of P ( t ) and a corresponding raster plot for that trial. The 100-trial raster for one run of the simulation was summated to form a PSTH. The close match between the simulated PSTHs (black) and the real PSTHs (red lines, reproduced from Figure 2 C and 2 E) indicates that the real responses to natural stimuli could be explained by neuronal selectivity to velocity features. The Pearson correlation coefficients between the predicted and observed PSTHs were 0.94 for the first-order neuron and 0.84 for the cortical cluster. Because these correlation values fall within the range obtained by comparing two real PSTHs generated from separate sets of 50 trials, we conclude that simulated spike trains are as similar to real spike trains as real spike trains are to each other. Thus, the velocity feature extraction properties of the neurons are sufficient to explain texture responses. Discussion Texture coding appears to derive from two fundamental processes: First, the whisker transmits a “kinetic signature” of the palpated surface to the receptors in the follicle. Second, the first-order neurons relay to the whisker region of cortex (through intervening stations) precise information about the kinetic features transmitted to the follicle. One kinetic feature is the “equivalent noise level”: Spike counts per whisk both for first-order neurons and for cortical neurons are proportional to the equivalent noise level of the texture-induced vibration. Thus, when texture vibrations differ in this quantity, neuronal spike counts also differ and thereby carry information that could, by itself, separate the textures. By the same token, when texture vibrations have similar equivalent noise levels, spike counts per whisk appear not to carry sufficient information. The second kinetic feature, then, is the temporal sequence of velocity events—distinctive velocity profiles induce distinctive temporal patterns in the spike trains with spike alignment of better than 0.2 ms in the first-order neurons and a few ms in the cortex. The stimulus playback method used here might not produce the identical input to the sensory receptor as occurs during active whisking [ 18 ]. The optical sensor at the base of the whisker did not register the bending nor tension (pulling) of the whisker. Moreover, active muscle contractions might affect sensory receptors at the interface between the whisker shaft and the inner membrane of the follicle. Thus, additional information related to surface texture might be available to the sensory system. In the present dataset, even after the possible loss of some texture-dependent information due to passive stimulation, the neuronal responses afforded a high degree of discriminability. We interpret the dataset as showing that vibration patterns, by themselves, must be a fundamental feature supporting the neuronal coding for texture. This awaits confirmation in experiments in actively whisking rats. A classical approach to investigating sensory coding is to map the relationship between well-controlled artificial sensory stimuli and evoked neuronal activity. This can provide a complete description of neuronal feature extraction properties [ 24 , 25 ], but it sheds little light on the brain activity underlying normal perceptual experiences. Moreover, the processing mechanisms that have evolved to extract behaviorally relevant information may operate inefficiently during artificial stimulation [ 26 ]. Another approach [ 27 , 28 ] is to measure neuronal activity during natural stimuli (i.e., visual scenes or animal calls). Here, the drawback is that the features evoking spikes during natural stimulation can be multidimensional, complex, and difficult to quantify, offering only limited insight into sensory processing mechanisms [ 27 ]. In principle, one can bridge the gap between artificial and natural stimuli by (i) measuring neuronal activity during ecologically relevant stimuli, stimuli that are collected from an animal's normal interaction with the environment, (ii) constructing tuning curves under artificial stimulation (usually white noise), and (iii) applying the tuning curves to the natural stimuli to test whether they account for the observed response. Because artificial stimuli only partially cover dimensions of stimulus space present in natural stimuli, and because of neuronal nonlinearity, this procedure typically provides neuronal simulations that match real neuronal output with a correlation of less than 0.5 [ 27 , 29 , 30 ]. Yet, the present experiments followed this same procedure and generated simulated PSTHs that were highly correlated with real responses (first-order neuron, 0.94; cortex, 0.84), approaching the upper bound set by the trial-to-trial variability that limits the correlation even between two real PSTHs. The simulations were successful for two reasons. First, recent work [ 11 , 12 ] uncovered the critical physical dimension—velocity—encoded by neurons. Second, receptor and cortical stimulus integration is linear to a first approximation—ongoing responses depend upon an integration process where preceding events affect the neurons independently of one another. As an alternative to the temporal model outlined here, a spatial model for texture coding has recently been proposed. It begins from the observation that whisker length varies systematically across the anterior-posterior dimension of the rat's snout [ 31 ]. When stimulated near the distal end, the posterior whiskers resonate at lower frequencies than do the shorter, anterior whiskers [ 32 ]. If different textures cause systematically different vibration frequencies, there might be texture-specific differences in the magnitude of vibration of posterior versus anterior whiskers [ 33 ]. Because whisker position is relayed in a somatotopic manner to the cerebral cortex [ 7 ], texture could be encoded by the location of the focus of activity in barrel cortex, much like frequency is encoded by the tonotopic organization of the cochlea and the auditory cortex. The resonance frequency hypothesis predicts that rats would fail to discriminate between surfaces using just a single whisker, yet they have been shown to perform texture discriminations after progressive clippings down to one or two whiskers [ 34 ]. Our results help explain the behavioral findings by emphasizing that, although additional information might be available due to differences in the mechanical properties among whiskers, even a single whisker can transmit large amounts of texture-specific information to its central neural circuits. This occurs because of the match between the feature of the whisker output signal that best distinguishes one texture from another (the kinetic signature of the vibration) and the tuning properties of first-order neurons. Cortical neurons conserve this kinetic signature in their firing patterns. Materials and Methods Recording the texture library and construction of the stimulus set Experiments were conducted in accordance with NIH and institutional standards for the care and use of animals in research. Subjects were ten adult male 250–350-g Wistar rats. In one set of anesthetized rats (urethane, 1.5 g/kg), “electrical whisking” [ 17 , 18 ] was generated by stimulating the right facial nerve (see Figure 1 A) with 1–2-V pulses of 100 μsec at 200 Hz for 60 ms to produce whisker protraction, followed by a passive 65-ms whisker retraction. For a selected whisker, horizontal and vertical movements at the base were registered by a two-channel optical sensor, each channel consisting of an LED light source and phototransistor ( Video S1 ). The two voltage signals were digitized (7,634 samples per second). Whisker movement was studied for 10 min under each of six conditions (see Figure 1 B). The angle traversed at the whisker base, averaged across all trials and all textures, was 25 degrees (SD ± 1 degree). Average translation across the edge of the textured surface was 3.08 mm (SD ± 0.14 mm). For each texture, a 50-s continuous record was extracted and sliced into 100 trials of 500 ms, each trial composed of two-dimensional position signals across four 125-ms whisks. For free whisks, a 250-s record was sliced into 500 unique trials. The stimulus set was constructed by splicing trials together at the point of maximum retraction (V H = 0), avoiding the introduction of any position or velocity discontinuity. A free whisk trial always separated two successive texture trials. A block was composed of five different texture trials (t 1–5 ) with free whisk trials (fw) interspersed, e.g., fw-t 3 -fw-t 5 -fw-t 1 -fw-t 2 -fw-t 4 . Before stimulus delivery, signals were low-pass filtered at 500 Hz. All data were manipulated in MATLAB software ( http://www.mathworks.com ). Analysis of vibration kinetics We sought to quantify the kinetic features that characterized each texture-induced vibration. We refer to the whisker trajectory in one dimension as x(t) and the whisker velocity profile as For each texture, we computed the spectrogram \| V(ω,τ n ) \| of the velocity profile where Δt is the 6-ms interval within which each spectrum was computed and τ n represents the series of N sequential time windows. Thus, each spectrogram (see Figure 1 , “velocity spectrogram” column) was composed of N = 906 spectra. Considering the duality between the direct and inverse Fourier space we can substitute Equation 1 into Equation 2 , to rewrite Equation 2 as Note that \|X ( ω,τ n )\| also corresponds to the spectrogram of the trajectory in position. Previous studies have shown that, when the stimulus set consists of sinusoidal whisker movements, the spike count per stimulus for cortical neurons is proportional to the product of the sinusoid's amplitude and frequency, Xω [ 11 , 12 ]. To test whether this coding principle extends to natural stimuli, we generalized the measure of Xω to the texture-induced vibration. X becomes a time-varying spectrum X(ω,t) where T is the entire time domain and Ω is the entire domain of frequency ω. This quantity—known as the “equivalent noise level”—represents the average amplitude of white noise velocity that dissipates the same average power as the signal of interest. It serves to characterize the entire kinetic signature by a single quantity, equivalent to mean value of the product Xω across all time intervals and all values of ω . Our experiments measured velocity in two dimensions, V H and V V . Equations 1–5 were generalized to a second dimension by adding, for each time window τ n , the separately measured spectrograms for the two dimensions. The “velocity spectrogram” column in Figure 1 B gives two-dimensional spectrograms as above. Similarly, in Figure 4 the equivalent noise level was calculated in two dimensions, averaged across 100 trials of each texture vibration, and then plotted in relation to neuronal spike count. Measurement of neuronal responses In a second set of urethane-anesthetized rats, neuronal recordings were made simultaneously from two sites. First-order neurons were recorded by advancing a single electrode (FHC, Bowdoinham, Maine, United States; http://www.fh-co.com ) into the right trigeminal ganglion according to stereotaxic coordinates. Cortical recordings were obtained by inserting a 100 microelectrode array (Cyberkinetics, Foxborough, Massachusetts, United States; http://www.cyberkineticsinc.com ) to a depth of 700–1,000 μm in the left barrel cortex [ 35 , 36 ]. Ganglion recordings were always single units, whereas cortical recordings consisted of a multiunit cluster at each channel. The principal whiskers (receptive field centers) in the first-order-only recordings were C3, E1, D6, E6, and γ (twice). The principal whiskers in the cortex-only recordings were A1, B4, E5, and E3 (twice). The principal whiskers in the paired first-order neuron-cortex recordings were δ (twice), E3 (twice), C2 (twice), and E4. Texture stimuli were delivered to a single whisker using a motor constructed from two orthogonal pairs of parallel piezoelectric wafers driven independently by horizontal and vertical signals ( Video S2 ). The whisker was inserted into a metal tube (0.33 mm inner diameter) with opening 1 mm from the skin. By optically monitoring the whisker shaft, we verified that movements precisely reproduced the previously recorded signals ( Figure S1 ). The second set of rats thus received whisker vibrations identical to those previously recorded during active whisking in the first set of rats. Construction of neuronal tuning curves and response simulations We selected V H and V V independently from a Gaussian distribution 7,634 times per second. The stimulus was then low-pass filtered (Chebyshev type II) at 500 Hz so as to not exceed the physical capacities of the piezoelectric wafer stimulator (instantaneous reversals of velocity cannot be achieved by any device). The noise stimulus was presented for 10 min after conclusion of the texture stimuli. We adopted a method of “forward correlation” between stimulus and response where, for all occurrences of a particular stimulus event, the ensuing neuronal spike trains were averaged to construct a response probability profile for that event. This required subdividing velocity space into discrete segments. In Figure 10 A, velocity space was partitioned into eight 45-degree angular sectors ( A 1–8 ) and ten radial sectors ( R 1–10 ). At each time point, angle corresponds to the direction of whisker movement, while radial distance corresponds to instantaneous speed. The positions of the radial boundaries were chosen to make each segment contain an equal number of instantaneous velocity events: Because velocity had a Gaussian distribution with a mean of zero, high-velocity events were less common, and consequently the radial boundaries were increasingly widely spaced as distance from zero increased. To test whether the complex, texture-induced spike patterns followed directly from the tuning curves, a more elaborate analysis was necessary. Each instantaneous velocity ( A,R ) during the whisk gave an ensuing spike probability profile. To simulate a spike train, a spike was generated in each time bin t (size 0.13 ms) with probability P ( t ), given by the average of the overlying spike probabilities associated with velocity events in the time window before t (time window was 1–2 ms before t for first-order neurons and 5–20 ms before t for cortical neurons). After completion of the simulation, P ( t ) was normalized so that the overall simulated spike count matched that in the real data; this normalization did not affect the temporal profile of the simulated PSTH. Supporting Information Figure S1 Comparison of Recorded and Played Back Whisking Trajectories The whisker movements presented as sensory stimuli were accurate reproductions of the movements recorded during electrical whisking in other rats. (A) Trajectories of two whisks (only horizontal channel shown) recorded during contact with sandpaper P280. (B) To test playback, a whisker was inserted in the piezoelectric motor guide tube and whisker displacements were measured by the optical sensor placed adjacent to the insertion point of the whisker. The trace shows recordings of the playback of the same two whisks of part A. (C) Magnified view of the traces indicated in the rectangle in (A) and (B). (161 KB PDF). Click here for additional data file. Video S1 Two-Dimensional Optic Sensor and Electrical Whisking Each optic sensor channel consisted of a pair of light tubes (one is a light source, the other leads to a photodiode). The two channels were mounted normal to each other in a metallic ring. Electrical stimulation of cranial nerve VII with 1-V pulses of 100 μs at 200 Hz for 60 ms produced whisker protraction that was followed by a passive 65-ms whisker retraction. The film shows four 125-ms free whisks in the air (8-Hz whisking), consisting of protraction (right to left) and retraction (left to right). Recorded at 500 frames per second with a Motion Scope 500 digital camera (Redlake, San Diego, California, United States; http://www.redlake.com ). Rat EW3, whisker C3. (9.1 MB MOV). Click here for additional data file. Video S2 Texture Playback with the Two-Dimensional Piezoelectric Motor The motor consists of two pairs of piezoelectric wafers with axes meeting at a mobile joint. The whisker to be stimulated was placed inside the metal tube, which is orthogonal to the joint. The film shows playback of four 125-ms free whisks in the air, consisting of protraction (left to right) and retraction (right to left). Recorded at 500 frames per second. (9.0 MB MOV). Click here for additional data file.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC544542.xml
552321	Malaria and urbanization in sub-Saharan Africa	There are already 40 cities in Africa with over 1 million inhabitants and the United Nations Environmental Programme estimates that by 2025 over 800 million people will live in urban areas. Recognizing that malaria control can improve the health of the vulnerable and remove a major obstacle to their economic development, the Malaria Knowledge Programme of the Liverpool School of Tropical Medicine and the Systemwide Initiative on Malaria and Agriculture convened a multi-sectoral technical consultation on urban malaria in Pretoria, South Africa from 2nd to 4th December, 2004. The aim of the meeting was to identify strategies for the assessment and control of urban malaria. This commentary reflects the discussions held during the meeting and aims to inform researchers and policy makers of the potential for containing and reversing the emerging problem of urban malaria.	Introduction Africa's population will almost triple by the year 2050. This expansion will occur primarily in urban areas and by 2025, 800 million people will live in urban communities. Especially affected will be West Africa, where the urban population annual growth rate of 6.3% is more than twice the rate of the total population growth. Today in the humid forest zone, more people live in cities than in rural areas and in twenty years time, two out of three West Africans will live in urban centres. While many of Africa's health problems are common to both urban and rural environments, recognizing and meeting the public health challenges in these growing cities is becoming increasingly urgent. Malaria has been considered a predominantly rural disease in Africa, primarily because suitable vector breeding sites are scarce in highly populated areas. Yet, although studies have shown that Anopheles mosquito breeding decreases with increasing proximity to the centre of urban areas [ 1 - 3 ], transmission of malaria still occurs. Clearly, the complex factors contributing to malaria risk in urban areas are not fully understood [ 3 ] but evidence is rapidly accumulating that the urban poor are at far higher risk from malaria than previously acknowledged [ 4 , 5 ]. The Malaria Knowledge Programme of the Liverpool School of Tropical Medicine and the International Water Management Institute/ Systemwide Initiative on Malaria and Agriculture convened a meeting in Pretoria 2 nd -4 th December 2004 to develop an evidence-based approach for evaluating and controlling urban malaria. Participants were drawn from seven sub-Saharan countries, Europe, North America and South Asia (see additional file). Recognizing the need for extensive cross-sectoral involvement and collaboration in dealing with the challenge of urban malaria, representatives from the research/ academic, NGO, development, policy-making and donor communities co-operated in the process to identify key knowledge gaps and opportunities for control. Included in the group were sociologists, clinical epidemiologists, entomologists and control specialists. Discussion Identifying the populations at risk in urban areas Urbanization is a recent phenomenon in Africa: in 1960 there were no African cities with one million inhabitants, today there are forty. Has malaria become a serious problem within these huge cities and their peri-urban environs? Data presented from studies in a number of sub-Saharan African cities (Brazzavile, Congo; Dakar, Senegal; Abidjan, Cote d'Ivoire; Cotonou, Benin; Ouagadougou, Burkina Faso; Dar es Salaam, Tanzania, and Accra and Kumasi, Ghana) showed clearly that malaria is a considerable urban health problem in Africa. The studies demonstrated great heterogeneities in malariometric indices both between and within cities. It was recognized that not only the major cities of Africa, but also many medium sized regional towns, home to a large proportion of the Africa population, have considerable levels of malaria [ 5 ]. With malaria risk unevenly distributed across urban environments, interventions must be preceded by the identification and prioritization of the most vulnerable. Vulnerability is not simply the result of low socio-economic status [ 6 ], although this is often a major contributory factor, but reflects factors beyond the individual level such as the proximity of the household to sites of urban agriculture or environmental/cultural factors working at the community level. Discussion focussed on research to define this risk, to improve access to correct diagnosis and appropriate treatment and effective preventative measures, and to identify accurate monitoring and evaluation tools tailored to the urban context. Prioritizing improved diagnosis and treatment for the vulnerable Misdiagnosis of malaria is a serious problem everywhere, but in areas of low malaria endemicity presumptive treatment of all fevers as malaria can result in over 75% of cases being misdiagnosed as malaria [ 7 ]. The effect of malaria misdiagnosis on the vulnerable will result in more ill health due to delayed diagnosis and repeat visits, overburdened health services, more severe malaria, loss of faith in health services, increase in real and perceived malaria resistance, chronic disease secondary to untreated infection, increased cost to patient and to health facilities and consistent misdiagnosis that will encourage detrimental health-seeking behaviour [ 7 ]. Effective provision of appropriate treatment also remains a serious challenge in urban settings. The Abuja Declaration stated that by 2005 "At least 60% of those suffering from malaria have prompt access to and are able to use correct, affordable and appropriate treatment within 24 hours of onset of symptoms." Despite the fact that access to quality health care is better on average in urban compared to rural zones, the formal public health facilities are often the last source of treatment used along the pathway to cure. Often malaria care initially involves leftover medicines from the home (from previously incomplete malaria or other treatment regimes), the purchase of cheaper herbal medicines or unprescribed conventional medicines. The problems of obtaining treatment from a health facility may be exacerbated by the need to obtain permission from an authority figure, absence from work and loss of income, the need to raise money to fund both the treatment and associated costs such as travel [ 6 ]. As a result, in Africa over 70% of malaria episodes in rural and over 50% in urban areas are self-diagnosed and self-treated [ 8 ]. With Home Management of Malaria proposed as an integral part of the Roll Back Malaria strategy, the consequences of presumptive treatment policies for malaria in the context of the introduction of newer and more expensive anti-malarial drug combinations urgently require further investigation [ 9 ]. Ensuring malaria prevention measures reach the vulnerable The highly focal nature of urban malaria requires targeting of interventions to specific urban districts, and therefore, requires detailed information on each area in advance. However, relationships between administrative boundaries, environment and population distribution are complex in urban areas, which makes them difficult to sample and characterize in a representative way. Strategies for population-representative sampling must incorporate a range of environments and populations to identify accurately environmental and other risk factors. This may be further complicated as urban populations can be highly mobile and in peri-urban areas there may be a high rate of turnover in groups of lower socio-economic status. Presentations from South, East and West Africa clearly demonstrated that Geographic Information Systems (GIS)-based approaches are valuable tools for assessing heterogeneities in risk factors for urban malaria, and for subsequent implementation and monitoring of interventions. Experienced researchers believe that the urban environment has advantages for the effective delivery of appropriate interventions. A number of studies have demonstrated that higher rates of coverage with insecticide-treated bednets can be achieved in urban areas [ 10 , 11 ], although whether or not the most vulnerable groups benefit, remains to be confirmed. Moreover, there is a growing realization within the commercial sector of the need to engage in health and broader social issues. The management of malaria can bring economic benefits to both businesses and the communities in which they operate. This has been powerfully demonstrated in two public-private partnership programmes in southern Africa that utilised indoor residual spraying to control malaria [ 12 , 13 ]. Larval control, achieved either by source reduction or larviciding, can be community directed and may be feasible in certain settings as part of a comprehensive, integrated vector management strategy. There is optimism in some communities about its efficacy and the results of further research into the costs and benefits of such interventions are awaited with interest. Environmental modifications may also be feasible if partners from the community and outside the health sector are engaged. Work from Sri Lanka has demonstrated how a very effective scheme to control malaria by modification of irrigation structures was accepted by the agricultural community because of the financial and water savings that the scheme introduced [ 14 ]. However, it was clear that obvious benefits from the intervention must exist to attract the involvement of non-health sectors. Conclusions The conclusions of the meeting have been summarised in the Pretoria Statement on Urban Malaria (Figure 1 ). While it is clear that urban malaria represents a major challenge for public health in Africa, the statement highlights that the unique nature of the urban environment provides an opportunity for malaria control. There are a number of reasons for this: the high population density in urban areas may facilitate increased coverage and impact of both interventions and health education programmes; the activities of departments in urban municipal authorities are typically better resourced and more easily mobilized than in rural areas; the extensive private health sector found in urban settings can be engaged to improve diagnosis, treatment and prevention of malaria. Solutions to the urban malaria problem must include groups from outside the health sector. The disease burden in the most vulnerable communities is a major obstacle to the economic growth of sub-Saharan countries and the challenge is to engage stakeholders at all levels in effective and sustainable intersectoral collaboration [ 15 ]. Urban malaria is uniquely amenable to prevention and control as the existing health, urban planning, agricultural and governance structures present opportunities for collaborative approaches that can include both the community and the substantial private sector. Figure 1 The Pretoria Statement on Urban Malaria. Authors' contributions All authors participated as session chairs in the technical workshop and were instrumental in producing the summary conclusions. All authors read and approved the final manuscript. Supplementary Material Additional File 1 List of attendees and affiliations Click here for file	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC552321.xml
544557	Identification of α-type subunits of the Xenopus 20S proteasome and analysis of their changes during the meiotic cell cycle	Background The 26S proteasome is the proteolytic machinery of the ubiquitin-dependent proteolytic system responsible for most of the regulated intracellular protein degradation in eukaryotic cells. Previously, we demonstrated meiotic cell cycle dependent phosphorylation of α4 subunit of the 26S proteasome. In this study, we analyzed the changes in the spotting pattern separated by 2-D gel electrophoresis of α subunits during Xenopus oocyte maturation. Results We identified cDNA for three α-type subunits (α1, α5 and α6) of Xenopus , then prepared antibodies specific for five subunits (α1, α3, α5, α6, and α7). With these antibodies and previously described monoclonal antibodies for subunits α2 and α4, modifications to all α-type subunits of the 26S proteasome during Xenopus meiotic maturation were examined by 2D-PAGE. More than one spot for all subunits except α7 was identified. Immunoblot analysis of 26S proteasomes purified from immature and mature oocytes showed a difference in the blots of α2 and α4, with an additional spot detected in the 26S proteasome from immature oocytes (in G2-phase). Conclusions Six of α-type subunits of the Xenopus 26S proteasome are modified in Xenopus immature oocytes and two subunits (α2 and α4) are modified meiotic cell cycle-dependently.	Background Eukaryotic cells, from yeast to human, contain large nonlysosomal proteases called proteasomes [ 1 ]. The 26S proteasome is part of the ubiquitin-dependent proteolytic system, which regulates proteins through a mechanism of selective degradation [ 2 - 4 ]. The 26S proteasome is composed of a 20S proteasome as a catalytic core and regulatory particles at either end. The subunits of the 20S proteasome subunits can be classified into two families, α and β. In eukaryotes, the 20S proteasome contains seven α-type subunits and seven β-type subunits. The fourteen kinds of subunits are arranged in four rings of seven subunits and form an α7β7β7α7 structure [ 5 ]. Fully grown frog oocytes arrest in the late G2 phase of meiosis. Maturation-inducing hormone (MIH) acts on the oocytes, inducing final maturation and triggering germinal vesicle breakdown (GVBD), and the oocytes arrest again at the second meiotic metaphase until fertilization. The proteasomes are thought to be involved in regulating the maturation and fertilization of oocytes [ 6 , 7 ]. Previously we identified the proteasomal subunit modified during oocyte maturation in Xenopus and goldfish as α 4 [ 8 , 9 ]. In the present study, we cloned three unidentified α-type subunits of Xenopus and prepared antibodies for a total of five subunits. Using a set of specific antibodies, we analyzed changes in all α subunits composing the 26S proteasome during the meiotic cell cycle. We demonstrated that 6 of the subunits exist as a heterogeneous population in frog oocytes and identified another subunit in addition to α4 which was modified meiotic cell cycle dependently. Results and discussion Isolation and characterization of cDNA clones A BLAST search of the Xenopus EST database was conducted using known proteasomal subunit α cDNAs. From the data for each subunit, full-length ORFs were obtained by PCR. The amplified cDNAs were 741, 726 and 786 bp long. The clones encode proteins of 246, 241 and 261 amino acid residues with a predicted molecular mass of 27463, 26402 and 29327 daltons, respectively (Fig. 1 ). Comparison of the amino acid sequence revealed that these molecules are highly homologous to the α1, α5 and α6 subunits in humans (overall identity 91.5–95.4%) [ 10 , 11 ], Drosophila (53.2–69.1%) [ 12 , 13 ] and yeast (53.2–61.7%) [ 14 - 16 ] (Fig. 2 ). Thus, we concluded that the cDNAs isolated in this study encode the α1, α5 and α6 subunits of the Xenopu s 20S proteasome. We named these clones α1_xl, α5_xl and α6_xl (α1, α5 and α6 subunits of Xenopus laevis ) according to a systematic nomenclature [ 5 ]. Figure 2 represents a comparison of amino acid sequences predicted from cDNA sequences of α-type subunits of the Xenopus 20S proteasome. Overall identity between the subunits was 25.1–38.4 %. A consensus sequence for α-type proteasomal subunits was conserved. Interestingly, a conserved sequence for β-type proteasomal subunits was found in the α3 subunit [ 17 ]. Figure 1 Amino acid sequence comparison of the Xenopus , human, Drosophila , and Yeast α 1 , α 5 and α 6 proteasome subunits. Amino acid sequence comparisons of α1 (A), α5 (B) and α6 (C) proteasome subunits are indicated. Matched sequences are boxed. Consensus sequences for calcium/calmodulin-dependent kinase II (CaMKII), cAMP/cGMP-dependent kinase (cAMP/cGMP), casein kinase II (CKII) and Ca 2+ -dependent kinase (PKC) are indicated. The numbers refer to the amino acid position at the end of each line. Figure 2 Amino acid sequence comparison of the Xenopus proteasomal α subunits. Matched sequences are boxed. The proteasomal α-type and β-type signatures were detemined by using the 'PROSITE' database [ 17 ] and are boxed. The numbers refer to the amino acid position at the end of each line. Comparison of proteasomes purified from immature and mature oocytes Polyclonal antibodies specific for five subunits (α1, α3, α5, α6, and α7) were raised against purified recombinant proteins. The specificity of the antibodies was examined by immunoblotting with the cytosol fraction and the purified 26S proteasome (Fig. 3 ). Each antibody preparation displayed a specific reaction for different polypeptides in both samples. Recombinant proteins from the cDNAs clearly cross-reacted with each antibody (data not shown). Thus, specific antibodies for each subunit were prepared. With these antibodies and previously described monoclonal antibodies for subunits α2 and α4 [ 18 ], changes to all α-type subunits during Xenopus meiotic maturation were analyzed. The modifications were demonstrated by 2D-PAGE (Fig. 4 ). The α7 subunit antibodies gave a single spot but all of the other antisera produced more than one spot, suggesting that the α1–α6 subunits undergo some type of modification in oocytes of Xenopus as demonstrated in other species [ 19 , 20 ]. A difference in the spots between the 26S proteasome from immature and mature oocytes was detected in the blots of subunits α2 and α4. In blots of α2 and α4, only a major spot was detected in the 26S proteasome from mature oocytes (in M-phase). It is suggested that the α4 subunit is phosphorylated in immature oocytes and dephosphorylated in mature oocytes [ 8 ]. Likewise, it is speculated that part of the α2 subunit is phosphorylated in interphase and dephosphorylated in metaphase. These results suggest that the subunits of 26S proteasomes are changed by meiotic cell cycle-dependent modifications. It can be speculated that these modifications are involved in the regulation of the meiotic cell cycle. Figure 3 Immunoblotting of the cytosol fraction and purified 26S proteasome. The cytosol fraction and purified 26S proteasome were electrophoresed under denaturing conditions (10.0% gel) and stained with Coomassie Brilliant Blue (CBBR), or immunostained with antibodies for α subunits of the 20S proteasome. Lanes cyt and 26S indicate the cytosol fraction and the 26S proteasome from immature oocytes, respectively. Molecular masses of standard proteins are indicated at the left. Protein bands of each subunit are indicated by arrows. Figure 4 2D-PAGE analysis of 26S proteasomes from immature and mature oocytes. The 26S proteasomes from immature (I) and mature (M) oocytes were subjected to 2D-PAGE followed by immunostaining with polyclonal antibodies against each of the Xenopus 20S proteasome subunits as indicated. The spots detected by each antibody are represented at high magnification and indicated by arrows. The spots differing between immature and mature oocytes are indicated by asterisks. The modification of proteasomal subunits and factors interacting with proteasomes may be involved in the regulation of proteasome function [ 21 ]. By two-dimensional polyacrylamide gel electrophoresis, up to 20 different polypeptides were separated from the 20S proteasome which was shown to be composed of 14 gene products [ 22 ]. Furthermore, changes in proteasomal subunit composition under different physiological conditions and the likely existence of a different subpopulation of proteasomes have been reported [ 12 , 23 ]. All these results suggest that the subunit composition of proteasomes, and likely their activity, is under complex control in vivo . Some of these changes may be due to post-translational modifications of the proteasomal subunits. Regarding protein modification, there have been several reports about the phosphorylation of proteasomal subunits. Phosphorylated proteasomal subunits were detected in crude preparations from cultured Drosophila cells [ 22 ]. Several subunits of the 20S proteasome could be phosphorylated in vitro by a cyclic AMP-dependent protein kinase copurifying with the bovine pituitary 20S proteasome [ 24 ]. Castaño et al. [ 25 ] (1996) identified the CKII phosphorylating subunit and its phosphorylation sites as the C8 component (α7 subunit) and serine-243 and serine-250, respectively. CKII was also reported to phosphorylate the C2 component (α6 subunit) in rice [ 26 ]. The phosphorylation of subunits in the 26S proteasome in vivo was investigated using cultured human cells. Mason et al. [ 27 ] (1996) showed the phosphorylated subunits to be the C8 (α7 subunit) and C9 (α3 subunit) components in the 20S core, and the S4 (Rpt2p) subunit and several other components in regulatory particles [ 28 ]. Recent approaches have revealed post-translational modifications to many of the subunits. In the yeast 20S proteasome, the α2- and α4-subunits are phosphorylated at either a serine or threonine residue, and the α7-subunit is phosphorylated at tyrosine residue(s) [ 20 ]. In the human 20S proteasome, more than two spots were identified in all α-type subunits except α5 and phosphorylation of the α7-subunit at serine-250 was revealed [ 19 ]. However the sites and kinases responsible for the phosphorylation of the α2 and α4 subunits of the 20S proteasome have yet to be demonstrated. The modification of these proteins is one possible mechanism regulating the functions of the 26S proteasome during the meiotic cell cycle. Consensus sequences for phosphorylation sites are conserved in these subunits [ 8 , 29 ]. Cyclic-AMP dependent protein kinase is responsible for the G2/M and metaphase/anaphase transitions [ 30 ]. Calcium/calmodulin-dependent protein kinase II is shown to be involved in the exit from metaphase II arrest at fertilization in Xenopus [ 31 ]. It can be hypothesized that these kinases are involved in the regulation of 26S proteasome activity. The identification of kinases and the phosphorylation sites of the α2 and α4 subunits may reveal how the modification of proteasomal subunits is involved in controlling the cell cycle. Currently, we have identified one of the protein kinase for α4 subunit as Casein KinaseIα [ 32 ]. Possible regulation of 26S proteasome activity by this kinase is under investigation. Recently, alternative subunits of proteasomes have been identified. In Drosophila where alternative α-type, β-type and 19S cap subunits are expressed from separate genes during spermatogenesis [ 33 ] and in Arabidopsis and rice where alternative isoforms of most proteasome subunits are differentially expressed from separate genes during development [ 34 , 35 ]. There are also examples of alternative β-type subunits in mammals (e.g., γ-interferon inducible "immunoproteasome" subunits β1i, β2i and β5i) [ 36 ]. Alternative subunits have yet to be identified in Xenopus , there is a possibility that the changes in the spots identified in this study may derive from differential expression of alternative subunits from paralogous genes. Conclusions (1) cDNAs for three α-type proteasome subunits ( α1_xl, α5_xl and α6_xl ) of X. laevis were identified. (2) Six subunits but not α7_XL are modified in immature oocytes in X. laevis . (2) α2, α4_XLs are modified during the meiotic cell cycle in X. laevis . Methods Purification of proteasomes Frogs ( Xenopus laevis ) were purchased from Jo-hoku Seibutsu Kyozai (Shizuoka, Japan) and maintained till used. 26S proteasomes were purified from immature oocytes and ovulated oocytes as described [ 37 ]. Electrophoresis and immunoblotting SDS-PAGE was carried out according to the method of Laemmli [ 38 ] (1970). 2D-PAGE (first dimension, NEPHGE; second dimension, SDS-PAGE) was carried out as described by O'Farrell et al. [ 39 ] (1977) using a precast polyacrylamide gel for NEPHGE (Immobiline Dry Strip pH3-10NL and pH6-11L for α4 subunit, Amersham biosciences) as reported [ 8 ]. Electroblotting and detection using antibodies were conducted as described [ 18 ]. cDNA cloning and sequencing Identification and sequence analysis of cDNAs. A BLAST search of the Xenopus EST database was conducted using known proteasomal subunit α cDNAs. From the data obtained for each subunit, independent sequences were linked and the full-length ORF sequences were eliminated(α1: BG347128 and CB558360, α5: BQ398972 and BJ043946, α6: BJ072624 and BJ091555). The specific primers for amplification of the full-length ORF were α1: 5'-GGAATTCCATATGTCTCGGGGATCTAGCGCG-3' and 5'-CCGCTCGAGGTCACGCTCAGCTAGTGCAAC-3', α5: 5'-GGAATTCCATATGTTCCTAACCCGCTCCGAG-3' and 5'-CCGCTCGAGGATGTCCTTAATAACTTCCTC-3', and α6: 5'-GGAATTCCATATGTTTCGCAATCAGTATG-3' and 5'-CCGCTCGAGGTGCTCCATAGGCTCCTCCTGC-3'), in which EcoRI (5'end) and XhoI (3'end) recognition sequence was added for cloning to the vector pET21a (Novagen). PCR was carried out using KOD DNA polymerase (TOYOBO) or LA taq DNA polymerase (TaKaRa), with Xenopus ovarian cDNA as a template, and the product was cloned to pET21a. The DNA sequencing was performed using a 377A DNA sequencer (Applied Biosystems). The sequences that include the full-length ORF identified here were deposited into GenBank (accession nos. AB164677, AB164678 and AB164679 for α1_xl, α5_xl and α6_xl , respectively). Pairwise comparisons of sequence homology were conducted using the Genetyx-Mac ver.12 computer program (Software Development, Tokyo, Japan). Production of recombinant proteins and preparation of antibodies The recombinant proteins were produced in E. coli BL21 (LysE) and purified by SDS-PAGE as described [ 6 ]. Polyclonal antibodies specific for each subunit were raised against purified recombinant proteins according to a procedure described before using guinea pigs [ 40 ]. Anti serums, which recognize the bands of each subunit, were obtained. Abbreviations bp, base pair; BLAST, basic local alignment search tool; cDNA, DNA complementary to RNA; EST, expressed sequence tags; kDa, kilodalton; NEPHGE, non-equilibrium pH gradient gel electrophoresis; PCR, polymerase chain reaction; SDS-PAGE SDS-polyacrylamide gel electrophoresis; 2D-PAGE, two-dimensional-PAGE. Authors' contributions YW carried out cDNA cloning, expression of recombinant proteins, antibody production and 2D-PAGE analysis. MT and RH participated in cDNA cloning, expression of recombinant proteins. YN and KI participated in coordination of the study. TT carried out the protein purification and also participated in the design of the study and drafted the manuscript.	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC544557.xml
552320	Cancer/testis antigens and gametogenesis: a review and "brain-storming" session	Genes expressed both in normal testis and in malignancies (Cancer/ Testis associated genes – CTA) have become the most extensively studied antigen group in the field of tumour immunology. Despite this, many fundamentally important questions remain unanswered: what is the connection between germ-cell specific genes and tumours? Is the expression of these genes yet another proof for the importance of genome destabilisation in the process of tumorigenesis?, or maybe activation of these genes is not quite random but instead related to some programme giving tumours a survival advantage? This review collates most of the recent information available about CTAs expression, function, and regulation. The data suggests a programme related to ontogenesis, mostly to gametogenesis. In the "brain-storming" part, facts in conflict with the hypothesis of random CTA gene activation are discussed. We propose a programme borrowed from organisms phylogenetically much older than humans, which existed before the differentiation of sexes. It is a programme that has served as a life cycle with prominent ploidy changes, and from which, as we know, the germ-cell ploidy cycle – meiosis – has evolved. Further work may show whether this hypothesis can lead to a novel anti-tumour strategy.	Introduction Cancer/Testis (CT) antigens are a group of tumour antigens with gene expression restricted to male germ cells in the testis and to various malignancies. Their function in tumours is enigmatic and a common between testis genes (gametogenesis) and cancer remains elusive. When the causal link is not evident, it is tempting to believe the association is random, and assign it to general aspects of "genome instability in cancer". However, we believe a more direct link may exist. As suggested in this review, possible clues may be found in the common evolutionary pathway between ploidy cycles in meiosis and polyploidy in tumour cells. The latter, along with CT antigen expression, is a characteristic feature of well-progressed tumours. However, before discussing such possibilities, it is necessary to review the established literature. The search for tumour antigens began in the 1960's with two groups identifying first alpha-fetoprotein (AFP), a serum marker for hepatoma and germ-cell tumours [ 1 ], and then carcinoembryonic antigen (CEA), a serum marker for colon and other epithelial cancers [ 2 ]. These antigens were discovered using heterogenous sera acquired by immunizing laboratory animals with human tumour material. However, only during the 90's did both cellular [ 3 , 4 ] and humoral [ 5 ] immune responses to human tumours get proper molecular definitions. The first CTA, MAGEA-1, was identified in 1991 by Boon and colleagues using T-cell epitope cloning, a very complicated and time-consuming method [ 3 ]. In 1995 the SEREX (serological expression cloning) technique to identify tumour antigens was developed by Pfreundschuh and colleagues [ 5 ], which remains the leading approach to identifying new antigens that elicit humoral immune responses. Besides MAGEA1, BAGE, and GAGE1 discovered by T-cell epitope cloning, SEREX very soon displayed more tumour antigens with a cancer/testis restricted expression profile (SSX2, NY-ESO-1, and SYCP-1). The term "cancer-testis (CT) antigen" was introduced by Chen et al. [ 6 ], who recognized this group of genes had little in common except their expression profile. By initial definition, expression of genes coding for CT antigens should be restricted in normal tissues to male germ cells in the testis and to malignancies of various histological types. However, the criteria proposed in the 90's are not true for all antigens of this group as seen today. Furthermore, for many of the recently discovered gene products with the described expression profile, no T-cell recognized epitopes have hitherto been identified. This is why CTA – "Cancer/Testis Associated" is a more appropriate name for this family of genes (and will be used in this context further in this review), because a lot of its members still need to be proven as possessing antigenic properties in cancer patients. Attributing genes to the CTA gene family is based on several characteristic features [ 7 , 8 ]: 1. Predominant expression in germ cells of the testis and generally not in other normal tissues. 2. Expression in a number of malignant tumours of various histological types. 3. Mapping of the gene to the X-chromosome 4. Membership of a multigene family. 5. Antigenic properties in tumour-bearing patients. Some exceptions to these criteria for certain CTAs will be described and discussed later. Expression of CTA genes To date, 89 individual CTA genes or isoforms have been described, which are organised in to 44 families (see additional file 1 ). From these, 19 families are testis-restricted, and 11 show additional expression in one or two somatic tissues. Nine are expressed in 3–6 tissue types besides testis, and 5 are ubiquitously somatically expressed. With the exception of the testis-restricted CTAs, the others also show expression in the pancreas but at levels as much as 10 × lower than in testis (based on mRNA expression levels from [ 9 ]). Expression of CTA was first shown in melanoma and all the classic CTA are expressed in this type of tumour, but since the 1980's, expression in various other tumours has been recognised ( additional file 2 ). The expression pattern of CTAs during spermatogenesis is of special interest. Functional analysis of these genes during gametogenesis might well give some clues about their possible role in tumours. Their expression is restricted exclusively to spermatogenic germ cells with other tubular cells (e.g. Leydig and Sertoli) being negative. This fits well with the findings of Yuasa et al. [ 10 ] who demonstrated that CTAs have much higher expression frequencies in the germ cell cancers (seminomas) than non-seminomas.. Different CTAs are expressed during different stages of spermatogenesis (Fig. 1 .), so one may imagine that their functions are versatile, starting from regulation of mitotic cycling in spermatogonia, association with the meiotic cycle in spermatocytes, and finalizing with acrosome maturation in sperm. Figure 1 Expression of CTAs during male germ cell development. References for Fig. 1: [25] [27] [28] [37] [40] [42] [53] [92] [93] [94] [95] [96] In normal tissues, expression of NY-ESO1, MAGE-A3, -A4, and -A8 through -A11 as well as of several members of the XAGE gene family is found in the placenta. NY-ESO1 and several XAGEs are also expressed in the fetal ovary [ 11 - 13 ]. Regulation of CTA expression The mechanisms involved in regulation of CTA expression have recently been comprehensively reviewed by Albert Zendman et al [ 14 ]. Thus, only a short recitation of some of the main points is provided here. Mostly, methylation processes are responsible for the ectopic derepression of CTA genes. Using the demethylating agent, 5-aza-2-deoxycytidine (5DC), expression of several CTAs in cultured tumour cell lines can be induced/upregulated [ 15 ]. 5DC entraps DNA methyltransferases in a complex with DNA, which leads to progressive loss of DNA methylation, thereby releasing transcriptional blockage. Such upregulated expression has been reported for several MAGE members – LAGE-1, SSX-2, CAGE, NY-ESO-1. For MAGE-A1, demethylation is necessary and sufficient for gene expression, suggesting demethylation is the primary mechanism of transcription control [ 16 ]. In this context, a recent discovery of a CTA, viz . 'Boris', is interesting. Boris is reported as being expressed in several types of malignancies, and normally plays a major role in regulating methylation processes during spermatogenesis – it removes imprinting from genes during the last mitotic division of spermatocytes (reviewed in [ 17 ]). Several lines of evidence indicate that expression of some CTAs is dependent not only on demethylation, but on other transcriptional mechanisms. Histone deacetylase (HDAC) inhibitors, on their own or in combination with 5DC, can also induce CTA expression, including MAGE, SSX, and NY-ESO-1 family members [ 15 ]. The CTA-rich region in Xp11.21-22 (e.g. SSX, MAGE-B) may escape X-chromosomal inactivation, but these genes are not normally expressed in females [ 18 ]. While global hypomethylation is common and prominent in colorectal cancer, few CTAs have ever been reported as expressed in this type of cancer [ 19 ]. Non-demethylation dependent induction of MAGE expression has been demonstrated by Park et al. [ 20 ], demonstrating that 40 mM NaCl induces the transcriptional and translational activation of MAGE-B1 and -B2 in specific tissues at hypertonic conditions. There exist definite expression patterns (sets) of different CTAs in certain tumours. Marked heterogeneity of CTA expression is found in cells of some tumours, which cannot easily be explained by a global demethylation process [ 21 - 24 ]. The mechanisms of ectopic transcriptional activation of CTA genes clearly needs more investigation. Function Information regarding the function and cellular localization of CTAs is far less comprehensive. Often the proposed function is based purely on sequence homology with another protein of a known function. The only CTA proteins functionally established in gametogenesis are SCP-1, involved in chromosome pairing during meiosis [ 7 ], OY-TES-1 which functions in acrosin packaging in the acrosome of sperm heads [ 25 ], SPO11 acting as a meiosis-specific endonuclease [ 26 ], and BORIS, which is involved in cancellation of imprinting by epigenetic reprogramming during the final round of mitosis in spermatogenesis [ 27 ]. However, contrary to the situation in meiosis where it is rapidly degraded after the meiotic prophase in spermatocytes, SCP-1 expression in tumours is not cell cycle restricted [ 7 ]. BORIS is a paralog of CTCF. CTCF is a highly versatile 11 zinc-finger factor involved in various aspects of gene regulation – X chromosome inactivation, reading of imprinting sites, etc. [ 27 ]. During spermatogenesis, Boris is expressed later than many other CTA genes ([ 27 ]; Fig. 1 .). Suggestions for the functions of other CTAs mostly arose from studies of their homology with some well-known proteins and their domains, TSP50 being protease-like, CT17 phospholipase-like, and CT15 metalloproteinase-like [ 28 ]. The CT15 gene encodes a disintegrin and metalloproteinase (ADAM) domain 2, which is a member of the ADAM protein family [ 29 ]. Members of this family are membrane-anchored proteins structurally related to snake venom disintegrins, which have been implicated in a variety of biologic processes involving cell-cell and cell-matrix interactions, including fertilization, muscle development, and neurogenesis. This member is a subunit of an integral sperm membrane glycoprotein called fertilin, which plays an important role in sperm-egg interactions [ 30 ]. It is a membrane metalloproteinase with a possible role in tumour evasion and metastasis. LDHC, the germ cell-specific member of the lactate dehydrogenase family, escapes from transcriptional repression, resulting in significant expression levels in virtually all tumour types tested. It might contribute to the constitutive activation of an anaerobic pathway in tumours, because its expression in tumours is not dependent on hypoxia [ 31 ]. Suggestions as to the role of MAGEs with a CTA expression profile mainly depend on studies of their ubiquitously expressed family members (e.g. Necdin, MAGE-D1, NRAGE, Dlxin-1). In general, the data point to a role for MAGEs via transcriptional regulation in cell cycle control and apoptosis. F.ex, Necdin-related MAGE proteins differentially interact with the E2F1 transcription factor and the p75 neurotrophin receptor [ 32 ]. The high level of homology among members of the MAGE family in both mouse and human suggests an important function both in testis and cancer. The products of the SSX genes belong to the family of highly homologous synovial sarcoma X (SSX) breakpoint proteins. These proteins may function as transcriptional repressors; SSX1, SSX2 and SSX4 genes have been involved in the t (X;18) translocation characteristically found in all synovial sarcomas [ 33 - 35 ]. This translocation results in the fusion of the synovial sarcoma translocation gene (SYT) on chromosome 18 to one of the SSX genes on chromosome X. The encoded hybrid proteins are probably responsible for transforming activity. In the nucleus of sarcoma cells, both diffuse and speckled localisations of SSX protein have been reported [ 36 , 37 ]. The HOM-TES-85 protein has structural peculiarities that are shared exclusively with the N-myc oncoprotein. However, functional studies are required for confirmation [ 38 ]. Besides, the C-myc proto-oncogene is a normal participant of spermatogenesis (Fig. 2 ). Figure 2 Oncogene expression during spermatogenesis (mainly in mice). From [70]. BRDT is similar to the RING3 protein family. It possesses 2 bromodomain motifs and a PEST sequence (a proline, glutamic acid, serine, and threonine cluster) characteristic of proteins that undergo rapid intracellular degradation). The bromodomain is found in proteins that regulate transcription [ 39 ]. PLU-1, a large multi-domain nuclear protein also has a strong transcriptional repression activity. It is a member of the ARID family of DNA-binding proteins. Plu-1 mRNA and PLU-1 protein are both highly expressed in the mitotic spermatogonia. The expression is reduced in the early prophase I stages (leptotene, zygotene), but reappears at pachytene, still being detectable in diplotene cells. It is located diffusely over the nucleus. PLU-1 might have a role in regulating meiotic transcription, restricted to certain meiotic stages [ 40 ]. The protein encoded by the SPANX gene targets the nucleus and associates with nuclear vacuoles and the redundant nuclear envelope in sperm cells [ 41 ]. In situ hybridization of human testis sections showed SPAN-X mRNA expression in round and elongated spermatids [ 42 ]. These redundant nuclear envelopes have a unique structure of limited chromatin sheets continued as annulate lamellae. Both these enigmatic structures have also been described in intact lymphomas [ 43 ] and irradiated lymphomas [ 44 ]. The protein encoded by IL13RA1 gene is a subunit of the interleukin-13 receptor. This subunit forms a receptor complex with IL-4 receptor alpha, a subunit shared by IL-13 and IL-4 receptors. This subunit serves as a primary IL-13-binding subunit of the IL-13 receptor, and may also be a component of IL-4 receptors. This protein binds tyrosine kinase TYK2, and thus may mediate the signalling processes that lead to the activation of JAK1, STAT3 and STAT6 induced by IL-13 and IL-4 [ 45 ]. SGY-1 (soggy-1), a secreted protein related to the Dickkopf protein family, is involved in suppressing the Wnt signal-transduction pathway controlling transcription activation of genes such as c-myc, c-jun, Fra, and cyclin D1 by preventing the accumulation of beta-catenin [ 46 , 47 ]. Wnt proteins are implicated in a wide variety of biologic processes including cell fate determination and patterning in early embryos, and in cell growth and/or differentiation in certain adult mammalian tissues [ 48 ]. Wnts can induce proliferation in different types of stem cells [ 49 , 50 ]. The importance of Wnt signalling during tumorigenesis has been recently emphasised [ 51 , 52 ]. NY-ESO-1 is one of the most immunogenic and therapeutically promising CTAs but functional studies on this gene are severely lagging behind its practical application. Unlike the majority of CTA genes, NY-ESO-1 stops its expression in well-progressed tumours, so it can be used as a marker to follow the early progression of testicular tumorigenesis [ 53 ]. CAGE [ 54 ] and HAGE [ 55 ] code for proteins with helicase-like features. Probably, it might be involved in recombination exchange in testis and recombination DNA repair in tumours. TPX-1 is now seen as an integral protein of the outer dense fibres and the acrosome of spermatids in rats [ 56 ]. In summary, we see that the functions of individual CTA, when known, are very diverse, including, for example, both activators and repressors of proliferation and transcription. Immunogenicity Immunogenicity in cancer patients is elicited only by short peptide sequences of CTA epitopes, which are presented on the tumour cell surface by HLA Class I molecules in the case of cytotoxic T lymphocyte (CTL) mediated immune responses and by HLA Class II molecules on the surface of APC (antigen presenting cells), in the case of T-helper cell (T H ) mediated immune responses. Identification of these epitopes is one of the main goals of CTA research. Knowledge of epitopes recognized by the immune system allows the creation of the tumour-specific vaccines. In the vaccination process, T-cell epitopes are often administered together with different adjuvants or cytokines, or delivered using peptide pulsed autologous dendritic cells, all of which are aimed at enhancing the immune reaction [ 57 ]. Currently efforts are being made to identify HLA class II restricted epitopes in order to promote T H responses, which are required to support the activity of CTLs, and provide a more "complete" immune response. However, vaccines designated to prime the immune system against tumours expressing various CTA have so far shown only partial clinical success [ 57 - 59 ]. Since the technique used to identify candidate tumour antigens has changed from T-cell epitope cloning [ 3 , 60 ], and SEREX (serological analysis of cDNA expression libraries) [ 5 , 61 , 62 ] to differential gene expression analysis by various techniques like RDA (representative difference analysis) [ 63 ], DD (differential display) [ 64 ] and SSH (suppression subtractive hybridization) [ 65 ] – and even further to bioinformatics – we cannot really be sure about the adequate use of the definition "antigen" in relation to CTA gene products. Whilst the former two techniques are dependent on the immunogenicity of specific epitopes in cancer patients, the latter ones are simply based on mRNA expression levels or even sequence homologies detected via search engines and provide no answer about immunoreactivity. In additional file 1 , one can see that an immune response is documented against only 19 of 44 CTAs. Immune recognition of the majority of these CTAs is cancer-related, occurring spontaneously in cancer patients, not in cancer-free individuals. The exceptions to this are humoral immune responses to MAGEB1/CT3.1 in systemic lupus erythematosus patients [ 66 ] and to SPA17/CT22 in vasectomised men [ 67 ]. CTA can be considered to be tumour specific. There exists the blood-testis barrier [ 68 ], which prevents the immune system from contacting with CTA gene products. Besides this, germinative cells do not express HLA class Ia molecules [ 69 ], so they cannot present their expressed proteins to the immune system. For these reasons, the immune system never comes into contact with these proteins and recognizes them as "non-self" structures. Brain-storming The function of most CTAs is unknown, although some role in regulation of gene expression (both activating and repressing) seems likely [ 8 ]. Are CTAs oncogenes? By definition, oncogenes are normal cellular genes participating in proliferation cascades, which are abnormally activated in tumours [ 70 ]. Therefore, with the exception of one (HOM-TES-85, which has structural homology with the myc-oncogene) CTAs can not formally be regarded as oncogenes. Are CTAs simply activated by the imbalanced genome, due to its instability in tumours? Some, such as SSX, whose ectopic expression is caused by the SYT-SSX fusion due to translocation t (X;18) in synovial sarcomas, certainly are. However, a large number of CTA genes are only located on the X-chromosome and this chromosome is not a specific site of chromosome breaks and translocations usually associated with tumours [ 71 ]. However, the chromosome region 20q13.2 containing the BORIS gene is commonly amplified or exhibits moderate gains of material in many human cancers. This has strengthened the idea that this region contains a major oncogene [ 72 , 73 ]. A preliminary report using RT-PCR has found that BORIS expression is detectable in over half of ~200 cancer cell lines studied, representing most of the major forms of human tumours, discussed in [ 17 ]. However, this observation awaits further confirmation. Theoretically, the aberrant expression of the amplified (by chance) Boris, which by analogy with CTCF may demethylate CTA genes located on X-chromosome, may in turn activate a large body of CTA genes in human malignancies. However, why are other gametogenesis-related CTA genes also activated from other chromosomes? It does not look like a chance event and therefore amplification of "Boris" may still not be random. Diversity of functions, including genes involved in ontogenesis suggest that CTAs are activated either as a result of the genome instability (however, why testicular genes?) or as part of a complex program. Also from this point, this looks like a program related to gametogenesis. The same idea was proposed by Old [ 8 ]. If CTAs generally do not enhance tumour growth except by stimulating proliferation, like oncogenes, another possibility is that they could do it by stimulating DNA repair or inhibiting apoptosis (combination of all three is possible). Indeed, some CTAs have relation to the DNA repair factors by homologous recombination. These are SPO11, SYCP1, helicase-like CAGE and HAGE acting in meiotic prophase of gametogenesis. In turn, homologous recombination in tumours was shown to act anti-apoptotically [ 74 ]. Are these CTAs restricted only to prophase of meiosis, where recombination takes place? It appears not. For example, some are associated with the spermatogonial stage (Plu-1 mRNA and PLU-1), and some with maturation of the acrosome (see Fig. 1 ). It is a program again, and why only male gametogenesis? But it also occurs in oogenesis [ 8 ]. Thus, prophase of meiosis (pairing and recombination) may be still the most important. So, ectopic activation of CTA genes is not entirely random, being induced from the sexual X-chromosome but also from loci on other chromosomes, relating to gametogenesis. However, it is curious then that the very common proto-oncogenes of proliferatives cascades also participate in gametogenesis, e.g. myc, ras, jun, etc. This can be seen in Fig 2 , (taken from the Janis Erenpreiss [ 70 ]). He and others postulated a link between gametogenesis and cancerogenesis before CTAs were revealed [ 75 - 77 ]. In turn, Old [ 8 ] looked at the problem from a new angle and suggested that CTAs provide a causal link between gametogenesis and cancer. This seems plausable, but does this illegitimate program in tumours embrace, even mosaically, only gametogenesis? May be only the DNA recombination repair component is common? But the CTA genes are repressing a wide Wnt family, which also functions in early embryogenesis. The program sounds more like an ontogenetic (life-cycle) one. Let us remember the experiments by Mintz and Illmeisee [ 19 ] who cloned normal genetic mosaic mice by introducting the nuclei of malignant teratocarcinoma into enucleated eggs. Both gametic, parthenogenetic and trophoblastic theories of cancer have also been proposed in the past [ 8 , 70 ] and can be viewed as embryonal or ontogenetic theories of cancer. Another question is whether CTA genes govern a life-cycle-like program with its key events similar to meiosis? In turn, to which process does meiosis provide a key – recombination and reduction division? Are tumours capable of these? We also have to consider why X-chromosomes are involved, and not Y-chromosomes, which are responsible for sex? Perhaps this is because the X-chromosome is better conserved evolutionary and appeared prior to sex discrimination. So, this looks like an ontogenetic program, which evolutionary preceded sexual (amphimictic) life-cycle, with the key in recombination and reduction division; but what is it? The answer is evident – ancient ploidy cycles [ 78 , 79 ]. In general, ploidy cycles are displayed as a cyclic increase and reduction of ploidy (chromosome number), and these involve the pairing of homologous chromosomes and their segregation, omitting one round of DNA replication. There may be reduction of ploidy 2n-n (sexual meiosis) or from nx to 2n DNA numbers, in the asexual ploidy cycles characteristic for a few protists, as in Amoebae and foraminiferans [ 80 , 81 ] as well as part of ontogenetic programs in the more developed taxons [ 82 , 83 ]. Contrary to gametic reduction in sexual meiosis, the reduction of ploidy in polyploid somatic cells is called "somatic reduction". In this context, expression of some CTAs by the placenta is of interest because trophoblast and decidua are the only mammalian tissue capable of endoreduplication creating enormously large ploidies [ 84 ]. an ability shared by many tumour cells. The latest studies on a silver fox revealed somatic reduction in the giant cells of a trophoblast [ 85 ]. Now let us return to DNA recombination repair in tumours as a means of survival, as already mentioned. Repair by homologous recombination can protect malignant tumour cells from apoptosis. In particular, as shown in our laboratory, endopolyploid cells employ this mechanism [ 86 ]. Likewise, expression of CTAs – endopolyploidy – is a hallmark of malignant tumour progression where there is deficient TP53 function [ 87 ]. In turn, some giant tumour cells show the capability to segregate their genomes and return to mitosis ([ 86 , 88 - 91 ], see Fig. 3 ). Figure 3 Reductional mitotic divisions generating low ploidy cells from one large polyploid cell. Non-treated Burkitt's lymphoma cell line, DNA staining with Toluidine-blue. × 2,500. Conclusion A hypothesis is put forward that activation of at least some of the CTA genes in p53-deficient human tumours could be due to the genetic program running "relic" ploidy cycles in tumour cells. This hypothesis offers new opportunities for the design of novel tumour treatment strategies. In particular, passive therapy using CTAs to prime the host immune system against the tumour could be replaced with gene therapy aimed to block the function of CTA gene products or even their expression. This approach is promising, because normally only the germ cells in the testis express CTA genes and they are well protected by the blood-testis barrier. Thus, there should not be any problem with tumour-specific priming, and respectively we would predict there to be no side effects. Supplementary Material Additional File 1 "Chromosomal localization, type of immune response, identification method and identification references for all known 44 CTA gene families". Click here for file Additional File 2 " CTA Frequency (%) of expression in various tumour types". Click here for file	/Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC552320.xml

Subsets and Splits

No community queries yet

The top public SQL queries from the community will appear here once available.