MIT Research News' Journal
[Most Recent Entries]
[Calendar View]
Thursday, January 28th, 2016
| Time |
Event |
| 12:00a |
A cleaner ballot box An effort to clean up local elections in Brazil has yielded new evidence about the prevalence of “voter buying” in one of the world’s largest democracies. A study co-authored by an MIT political scientist finds that audits of voters reduce the electorate by 12 percentage points in local elections, and lower the chances of mayoral re-election by 18 percentage points. Fraudulent electoral practices are also much more prevalent in small towns than in larger places.
While patronage systems are often thought to materially reward existing residents for their votes, such political machines not only “influence actions of the electorate,” as the new paper states; they also change the composition of the electorate.
In Brazil, contemporary procedures and technology have left political machines unable to invent voters and votes out of thin air. Instead, the electoral subterfuge in question consists of efforts to “import” real voters into places where they do not legally belong, through dubious registration practices.
“In Brazil, where they have a good voting system and there is essentially no ballot stuffing, you can still have this other vector for fraud that can have quite large effects on who wins and loses,” says F. Daniel Hidalgo, an assistant professor in MIT’s Department of Political Science and a co-author of the paper. “In large swaths of the territory, it’s a big issue.”
Yet the results also suggest the limitations of “electoral tourism,” as it is sometimes called globally. Compared to old-fashioned ballot-stuffing practices, this form of fraud is more expensive, and thus may be more limited in scope. That may explain why it seems more common in smaller places; importing a given number of voters would have a lesser impact in a larger municipality.
“This is a good story, in the sense that cheaper ways of rigging the vote are unavailable,” Hidalgo says.
The paper, “Voter Buying: Shaping the Electorate through Clientelism” has been published by the American Journal of Political Science. The authors are Hidalgo and Simeon Nichter, an assistant professor of political science at the University of California, San Diego.
The 80 percent solution
The study uses what social scientists call a “regression discontinuity design” to arrive at its findings. Regression discontinuity designs usually compare social outcomes that fall narrowly on opposite sides of some sort of cutoff, such as one determined by a policy decision.
In this case, Brazilian electoral courts order audits of voting rolls when the eligible electorate exceeds 80 percent of a municipality’s population. So, by comparing the election results in places where just under 80 percent of the population was eligible to vote, with results in places that triggered audits because slightly more than 80 percent of the population could vote, Hidalgo and Nichter were able to see auditing-related differences between places that were otherwise broadly similar in voting terms.
“Because it’s fraud, it’s hard to measure,” Hidalgo observes. In Brazil, he adds, “There are a lot of anecdotes, but it hasn’t really been studied systematically.”
In 2007, about one-quarter of Brazil’s 5,564 municipalities exceeded the 80 percent threshold, although a slightly smaller fraction was actually audited. The researchers based the study on a wave of electoral audits conducted in 2007 and 2008, the largest set of such checks Brazil has conducted in decades.
About three-quarters of the towns with suspiciously high voter registration numbers had a population of 11,300 citizens or fewer; only two had populations over 90,000.
A level playing field for challengers
Cleaning up elections has a disproportional effect on incumbents, Hidalgo and Nichter write, because incumbents are likely to have “greater access to resources” than challengers do, which enables to them conduct more voter buying. Those resources can include access to public funds or social programs, or staff that process registration transfers. In all, the proportion of incumbents winning re-election after audits took place dropped from 53 percent to 35 percent.
“Mayoral elections in Brazil are incredibly competitive,” says Hidalgo, noting that a paradox of sorts exists in the country: “It’s a vibrant democracy, but that generates strong incentives for mayors to do this kind of thing [voter buying].”
Other scholars regard the work as innovative and interesting. Haracio Larreguy, an assistant professor of government at Harvard University, who has read the paper, praises the study as providing “the first estimates of the extent of this issue in a developing country, which is a context where voter buying is believed to be of big relevance.” Larreguy also states that the research “opens the door to more work on voter buying in other countries.”
Despite the results, Hidalgo emphasizes that in his view, Brazil’s electoral system, in all its aspects, generally functions well.
“I don’t want to overstate and say this undermines the electoral system more broadly,” he says, adding: “In some ways I would say their electoral system is better than ours [in the U.S. and Western Europe],” due to the presence of electronic voting, the use of nonelected officials to run elections, and other factors.
Still, Hidalgo adds, he thinks studies of electoral malfeasance are always worth conducting.
“I’m especially interested in interventions that increase participation, reduce fraud, and clean up elections,” Hidalgo says. | | 12:00a |
Faster drug discovery? For pharmaceutical firms, gene-expression profiling has become a valuable tool for drug discovery. This process involves measuring the activity of a cell’s genes in response to drugs, to determine the compounds’ effectiveness, toxicity, and other characteristics. Conventional profiling methods, however, are inefficient or expensive, sometimes costing millions of dollars.
Now Genometry has commercialized a high-throughput gene-expression assay developed at the Broad Institute of MIT and Harvard, which operates at a fraction of the cost of conventional methods. It does so by using measurements of 1,000 genes to accurately and quickly estimate the activity of all the 20,000 or so genes expressed in a cell.
The fast, low-cost assay allows for much larger experiments than previously possible, and for gene-expression profiling to be used much earlier in the drug-discovery process — which could speed things up, says Genometry co-founder, president, and CEO Justin Lamb, a former Broad Institute researcher. “Rather than profiling only a handful of compounds at the end of the search for a new drug to confirm that you got what you wanted, you can use profiling right at the start of the search, and hence do the search in much more efficient ways, because you have much more information,” he says.
A dozen pharmaceutical firms and other companies are now using the assay. Last October, Genometry signed a multiyear contract with Janssen Pharmaceutica to generate gene-expression profiles for 250,000 compounds in the Belgian firm’s small-molecule library. This represents the first time gene-expression profiling has been applied at such a large scale. The data will be used for drug screening and improving the selection of candidate drugs before clinical studies.
Lamb co-developed the assay, called L1000, with researchers in Broad Institute Chief Scientific Officer Todd Golub’s group, including Genometry co-founder Aravind Subramanian and David Peck, both currently researchers at the Broad.
Landmark genes
For gene-expression profiling, researchers have traditionally used microarrays, pieces of DNA arranged on silicon wafers, or polymerase chain reaction (PCR), which copies DNA fragments in test tubes. PCR is more accurate, but microarrays are faster: It takes the same amount of time to measure a few dozen genes using PCR as it does to measure an entire transcriptome — the full range of genes expressed in a cell —using microarrays.
But these experiments, Subramanian says, can cost up to $500 to generate the cellular response to a single drug or other “perturbagen.” Matching signatures against libraries of a million or so drug candidates and different cell types, adds up.
L1000 takes advantage of the fact “that genes don’t act independently of each other. They travel in clusters,” according to Subramanian.
Instead of recording all the gene expression in a cell, L1000 measures the expression of approximately 1,000 so-called landmark genes that have been selected because they have special qualities, such as being minimally redundant across the genome or widely expressed in different cell types. According to Subramanian, because of these properties, the landmark genes together contain around 80 percent of the information in the entire transcriptome.
Genometry provides L1000 as a service. Clients send lysate — a fluid of broken-apart cells — in 384-well plates, usually dozens at a time, to Genometry’s headquarters and lab in Kendall Square. After some initial processing, the samples are mixed with microscopic beads of different colors coated with DNA from landmark genes, before being run through a version of a flow cytometer. Varying fluorescence intensity of each bead denotes varying degrees of each gene’s expression. Based on those measurements, a computational inference model infers how all other genes will behave. With fewer measurements, throughput increases and the price drops to a few dollars per sample, compared with $500 by other methods.
Pharmaceutical firms will use L1000 data from potentially hundred of thousands of compounds as an initial screen, or to characterize their entire library of chemicals, as is the case with Janssen. Firms could also use the data for “hit selection” to narrow down the number of viable candidate drugs, Lamb says. A firm may have, say, 1,000 compounds to potentially treat a certain disease. Running those through L1000 will quickly reduce that number to perhaps 50 of most selective compounds and the ones with the fewest off-target effects. “It’s a convenient way to prioritize compounds for drug development,” Lamb says.
“Google” for drug discovery
In 2012, Lamb and Subramanian formed Genometry to commercialize L1000, only after it had proven its mettle at Broad Institute. “The purpose of [Genometry] was not so much to take a hand-wavy academic idea and make it into a product — it was to take a tried, tested, and proven technology and come up with a mechanism by which the outside world can get easy and effective access to that innovation,” Lamb says.
L1000 was developed for — and has become an integral component of — an ongoing Broad Institute project called the Connectivity Map, which aims to become, essentially, “Google for drug discovery,” according to the project’s website.
The Connectivity Map, launched in 2005, is a collection of gene-expression responses to perturbagens, including chemical and genetic manipulations. “If you describe the actions of drugs, the effects of genetic manipulation, and disease states in a common language, connecting a disease with a potential remedy, for example, is a relatively easy pattern-matching problem. Gene expression provides a particularly good vocabulary for this,” Lamb explains.
Users of the Connectivity Map, for instance, can upload a list of genes whose expression pattern characterizes a disease and, with one click, receive a list of compounds ranked by their effect on those genes. It’s similar to using keywords in Google to call up pages ranked by their relevance, Subramanian says.
At the Broad Institute, L1000 has now helped researchers archive profiles for around 5,000 drugs — about half the drugs ever tested in humans — and perturbation of 3,000 genes using CRISPR genome editing and other tools. The Broad Institute hopes to amass data on the connections among all drugs, genes, and diseases in the near future, Subramanian says.
Lamb says Genometry aims to help pharmaceutical firms use gene-expression data to effectively digitize their compound collections. In doing so, firms can use various tools to “search for compounds with desirable characteristics, determine if new chemical matter will be effective against various diseases, or if they are too similar to existing drugs to be commercially viable,” he says.
“It’s important to not just digitize libraries available to academics,” Subramanian adds, “but to also make sure companies are achieving the impact they want.” | | 1:59p |
How severe maternal inflammation can lead to autism-like behavior In 2010, a large study in Denmark found that women who suffered an infection severe enough to require hospitalization while pregnant were much more likely to have a child with autism (even though the overall risk of delivering a child with autism remained low).
Now research from MIT, the University of Massachusetts Medical School, the University of Colorado, and New York University Langone Medical Center reveals a possible mechanism for how this occurs. In a study of mice, the researchers found that immune cells activated in the mother during severe inflammation produce an immune effector molecule called IL-17 that appears to interfere with brain development.
The researchers also found that blocking this signal could restore normal behavior and brain structure.
“In the mice, we could treat the mother with antibodies that block IL-17 after inflammation had set in, and that could ameliorate some of the behavioral symptoms that were observed in the offspring. However, we don’t know yet how much of that could be translated into humans,” says Gloria Choi, an assistant professor of brain and cognitive sciences, a member of MIT’s McGovern Institute for Brain Research, and the lead author of the study, which appears in the Jan. 28 online edition of Science.
Finding the link
In the 2010 study, which included all children born in Denmark between 1980 and 2005, severe infections (requiring hospitalization) that correlated with autism risk included influenza, viral gastroenteritis, and urinary tract infections. Severe viral infections during the first trimester translated to a threefold risk for autism, and serious bacterial infections during the second trimester were linked with a 1.5-fold increase in risk.
Choi and her husband, Jun Huh, were graduate students at Caltech when they first heard about this study during a lecture by Caltech professor emeritus Paul Patterson, who had discovered that an immune signaling molecule called IL-6 plays a role in the link between infection and autism-like behaviors in rodents.
Huh, now an assistant professor at the University of Massachusetts Medical School and one of the paper’s senior authors, was studying immune cells called Th17 cells, which are well known for contributing to autoimmune disorders such as multiple sclerosis, inflammatory bowel diseases, and rheumatoid arthritis. He knew that Th17 cells are activated by IL-6, so he wondered if these cells might also be involved in cases of animal models of autism associated with maternal infection.
“We wanted to find the link,” Choi says. “How do you go all the way from the immune system in the mother to the child’s brain?”
Choi and Huh launched the study as postdocs at New York University School of Medicine, working with Dan Littman, a professor of molecular immunology at NYU and one of the paper’s senior authors. They began by injecting pregnant mice with a synthetic analog of double-stranded RNA, which activates the immune system in a similar way to viruses.
Confirming the results of previous studies in mice, the researchers found behavioral abnormalities in the offspring of the infected mothers, including deficits in sociability, repetitive behaviors, and abnormal communication. They then disabled Th17 cells in the mothers before inducing inflammation and found that the offspring mice did not show those behavioral abnormalities. The abnormalities also disappeared when the researchers gave the infected mothers an antibody that blocks IL-17, which is produced by Th17 cells.
The researchers next asked how IL-17 might affect the developing fetus. They found that brain cells in the fetuses of mothers experiencing inflammation express receptors for IL-17, and they believe that exposure to the chemical provokes cells to produce even more receptors for IL-17, amplifying its effects.
In the developing mice, the researchers found irregularities in the normally well-defined layers of cells in the brain’s cortex, where most cognition and sensory processing take place. These patches of irregular structure appeared in approximately the same cortical regions in all of the affected offspring, but they did not occur when the mothers’ Th17 cells were blocked.
Disorganized cortical layers have also been found in studies of human patients with autism.
Preventing autism
The researchers are now investigating whether and how these cortical patches produce the behavioral abnormalities seen in the offspring.
“We’ve shown correlation between these cortical patches and behavioral abnormalities, but we don’t know whether the cortical patches actually are responsible for the behavioral abnormalities,” Choi says. “And if it is responsible, what is being dysregulated within this patch to produce this behavior?”
The researchers hope their work may lead to a way to reduce the chances of autism developing in the children of women who experience severe infections during pregnancy. They also plan to investigate whether genetic makeup influences mice’s susceptibility to maternal inflammation, because autism is known to have a very strong genetic component.
Charles Hoeffer, a professor of integrative physiology at the University of Colorado, is a senior author of the paper, and other authors include MIT postdoc Yeong Yim, NYU graduate student Helen Wong, UMass Medical School visiting scholars Sangdoo Kim and Hyunju Kim, and NYU postdoc Sangwon Kim. |
|