How to get negative data published in Science (get someone else to publish the opposite first!)

Poor Science. Nearly two years ago, they published a paper from Steve Reiner's lab proposing a new model for memory T cell differentiation. It was one of the most talked-about articles in immunology that year, and Science even named the story one of its "Breakthroughs of the Year" in 2007. Among immunologists, reactions ranged from "This is so elegant it must be true" to "This exists in such an alternate universe to all published data that it cannot be true," to "If this is true, why haven't I discovered it already?" Now, though, Science is obliged to publish whatever follow-up papers come along to try to resolve the controversy.

The idea is this: a naive T cell is like a stem cell. When it gets activated, it divides asymetrically such that one daughter cell maintains "stemness" and becomes a long-lived memory cell, and one daughter cell differentiates into an effector T cell. The memory cell sticks around, keeps its multipotency, and the effector cell takes care of all the immediate disesase-fighting tasks, then dies. The axis of cell division is set up by the interaction of the naive T cell with an antigen presenting cell (APC) and a specialized structure there called the "immunological synapse." One important function of the synapse, therefore, would be to indicate to recruit factors to one end of the T cell that would help it become the effector daughter. Asymmetrical division is a well-characterized mechanism for regulating stemness and differentiation in many developmental systems across phyla.

Models for memory T cell development can be broadly divided into two groups: divergent development, in which each daughter cell becomes either an effector or memory cell (based on stochastic events that may involve cytokine signaling or an unidentified coreceptor), and sequential development, in which every stimulated cell becomes an effector, and a subset of those later develop a memory phenotype. Reiner's asymmetrical division hypothesis riffs on divergent development, suggesting that the signaling that determines whether a daughter is an effector or memory cell happens prior to cell division. So why was it so controversial? First, there was extant evidence from adoptive transfer experiments showing that effectors could differentiate into memory cells. Second, a large component of the data in that paper was FACS plots using cells that had been labeled with CFSE prior to stimulation. Dilution of the dye allowed the investigators to tell which cells had divided exactly once. Within that singly-divided population, there were two populations of cells with different levels of expression of a variety of markers, including CD8. This suggested that there was asymmetry in CD8 levels across the plane of cell division. It's an elegant assay. The problem was that immunologists do CFSE-dilution experiments ALL the time. There are probably tens of thousands of published plots of CFSE versus CD8, and nobody had ever noticed the bimodal CD8 distribution in the first division before. Neither, were they able to find them when they looked at their old experiments.

So, for 2 years, the field has been waiting for someone to replicate or refute Reiner's findings, and particularly to use a genetic system that might have a defect in asymmetric division, so that the FACS assay and memory outcome could be correlated. Two papers published in Science in January don't do this directly, but they were clearly written (and published) with an eye to the asymmetric division controversy. In fact, while the papers seem to be fine pieces of work, I can't imagine they'd warrant back-to-back publication of science had the Pandora's box of asymmetric division not been sprung open.

Ed Palmer's group (Teixeiro et al) made a mouse expressing a TCR with a single amino acid mutation in its CART domain. When naive T cells from this mouse were stimulated, the effector phase of the response was completely intact, with cells demonstrating normal proliferation, upregulation of CD62L, and expression of effector proteins like IFN-gamma, TNF, and granzyme. The memory phase, on the other hand, was entirely lacking. Specific mutant cells were virtually undetectalble in mice 2 weeks after immunization. Somehow, the authors found enough cells to do a battery of test on them, and found that the mutants failed to upregulate CD25, LFA-1, Granzyme B, and FasL upon restim. They also had defects in cytokine production and killing. Interestingly, on some of the parameters, the mutant cells catch up to wild type cells by day 4 after the restimulation. The authors conclude from this that the mutant cells have a failure of memory generation. An alternate hypothesis, however, is that the mutant cells have a failure in memory survival. The small cell numbers recovered could represent rare cells that didn't get stimulated the first time around, and the slow response to antigenic stimulation compared to wildtype could actually reflect the differences in kinetics of naive and memory responses. A parallel condition using naive cells would have been interesting to see.

The paper becomes relevant to the asymmetric division hypothesis because the authors saw a delay in TCR and PKC-theta polarization to the synape in mutant cells. This was not experimentally tied to cell division and differentiation, but it was correlated with failure of cells to translocate NF-KB to the nucleus. Interestingly, the examples they show of staining for NF-KB component cRel makes it look more like the protein failed to polarize than failed to translocate to the nucleus, but perhaps this was just poor image choice. So, do these correlated failures of polarity and memory support Reiner's hypothesis? Sort of. Or, they don't not support it. An argument could be made that the CART domain controls recruitment of polarity factors required for asymmetric division. This is somewhat refuted by a piece of data burried in the supplemental figures, showing that Scribble (a well-known polarity protein of probable importance in T cells based on work by Sarah Russel and Andy Chan) polarized perfectly in the CART mutant cells.

In the second paper, Bannard et al used a lineage tracking system to address the sequential model of memory development. They created a mouse in which a tamoxifen-inducible Cre is driven by the Granzyme B promoter. This means that cells have functional Cre in the nucleus only when tamoxifen is administered and the cells are sufficiently activated to express Granzyme B. the mice were then crossed to the ROSA26EYFP mouse, which expresses YFP when Cre has excised a stop element in the gene. This system allowed the researchers to irreversibly mark T cells that had undergone effector differentiation. The authors spent a lot of time in the paper validating their mouse in vitro and in vivo, and they were really interested in where and when the cells were dividing. The key experiment with regard to memory is a reinfection experiment toward the end of the paper. The authors infected their mice with influenza virus, waited 7 weeks, and then reinfected. 7 days later, they looked at the fraction of flu-specific cells that were YFP-positive, and found that YFP-positive cells expanded similarly to the total population. Their conclusion was that this demonstrates that effectors can differentiate into memory cells. Lesson: when Figure 3, the climax, of your Science paper starts with something like "Absence of impaired secondary clonal expansion..." you can bet that the reason your paper is in Science is that someone else published the wacky opposite first.

A few criticisms: first, if memory cells come from effectors, shouldn't the YFP-positive cells (the long-lived former effectors) expand more than the YFP-negative cells (which are presumably naive)? Second, this work tells us that effectors CAN become memory cells, but it doesn't by any stretch show that only effectors can differentiate into memory cells. At best, this keeps the sequential and the divergent development models both open. The approach that really would have differentiated between the models uses the following (somewhat complicated) mouse:

Transgene 1: CD25-driven ER-Cre
Transgene 2: Granzyme B-driven ER-Flp
Transgene 3: ROSA26-YFP (with stop element removed by Cre-Lox system)
Transgene 4: ROSA26-Cherry (with stop element removed by Flp-Frt system)

This would allow the experimentor to distinguish between cells that had never been activated (non-fluorescent), cells that had been activated but did not differentiate into effectors (YFP+), and cells that have been activated and fully differentiatied into effector cells (YFP+Cherry+).
If you then do a re-infection experiment, it should be possible to look at the relative populations of YFP+ and YFP+Cherry+ cells. If there is only sequential development of effectors and memory, there shouldn't be any YFP+ only memory cells, and all expansion should come from the double-positive and double-negative populaitons. If there is divergent development at the first cell division, there should be a big expansion of cherry only cells.

Now everyone get to work making and breeding that awful mouse.

In the end, it looks like the jury remains out on the Reiner model. The Teixeiro paper gives us the only other existing correlation between polarization and differentiation, but fails to connect these observations to cell division, a connection that is essential for Reiner's model. Particularly suspect is that they have a perfect system (high-functioning mutant) in which to use Reiner's assays to look at asymmetric division. That those experiments aren't shown suggests that the palmer lab is having as much trouble replicating Reiner's assays as everyone else I know. The Bannard paper argues against an asymmetric division model, but certainly doesn't rule it out. I guess there are likely to be a few more high-impact papers on this topic before we understand the real story.

References:

Chang JT et al. Asymmetric T lymphocyte division in the initiation of adaptive immune responses. Science. 2007 Mar 23;315(5819):1687-91.

Bannard O et al. Secondary replicative function of CD8+ T cells that had developed an effector phenotype. Science. 2009 Jan 23;323(5913):505-9.

Teixeiro E et al. Different T cell receptor signals determine CD8+ memory versus effector development. Science. 2009 Jan 23;323(5913):502-5.

Things we wish we'd known about a lot sooner: TLRs

Most scientists operate under the assumption that basic research is important for human health, even if its applications are not immediately apparent. A paper in the January issue of Nature Medicine demonstrates just how important an understanding of basic immunology can be, and in the process, solves a decades-old vaccine mystery. Delgado et al. take on the issue of why a vaccine against Respiratory Syncytial Virus (RSV) developed forty years ago not only failed to protect infants from the disease, but caused serious illness, and even death in some subjects.

The RSV vaccine consisted of wild-type RSV grown in primate cells, inactivated by formalin fixation, and mixed with alum as an adjuvant. By the time the trial in question started, the researchers had already carried out safety studies in four animal species and adult volunteers. Because RSV is primarily a disease of infants (around half of babies are infected in their first year,) a vaccine is not useful unless it can generate protective immunity in that population. Furthermore, virtually all potential adult volunteers would have already been exposed to the virus many times, obviating the testing or use of the vaccine in that population. The main efficacy study, therefore, took place in infants.

23 babies were given the vaccine, and on the whole, had a measurable increase in their neutralizing antibody titer (6x control) and a big, big increase in their complement-fixing antibody (30x control). The researchers concluded from this that the vaccine was immunogenic, but when the RSV season hit that winter, there was an unexpected result. The vaccinated subjects got far more severely ill when exposed to RSV (80% required hospitalization) and 2 died. Kim et al. did not suggest a mechanism for this result, but in their discussion they refer to something called the "paradoxical vaccine effect," which I've never heard of but apparently happened all the time back then, since they cite failed vaccines against rickettsia, trachoma, mycoplasma, and measles as examples. One has to wonder why they kept trying to make vaccines when the failure rate was so devastating, and whether, given todays safety standards, all of these trials would have ever been allowed.

To investigate the source of RSV's paradoxical vaccine effect, Delgado et al. immunized mice with a range of RSV preparations. One was inactivated by formalin, just like the human vaccine, another other was UV-inactivated, a third vaccine was made of purified protein, and the final preparation contained live replicating virus. When the immunized mice were re-challenged with live virus, only the mice previously exposed to live virus and mice who had received a placebo vaccine avoided enhanced respiratory disease (ERD). The authors went on to show that mice vaccinated with inactivated virus produced antibody with far lower affinity for model epitopes than the mice that received live virus.

Why did affinity maturation fail? First the authors look at draining lymph nodes following immunization and see attenuated upregulation of CD40, CD80, and Cd86 on dendritic cells, decreased CD4 T cell proliferation and cytokine production, and a failure to develop germinal centers. The authors argue that inactivated virus induces these components of immunity poorly because it fails to stimulate TLRs. To support this conclusion, the authors use MyD88 heterozygous mice, and demonstrate that they fail to generate high affinity antibody against live RSV. Whether MyD88 hets actually constitute a good model of reduced TLR activation is not clear, as the authors do no experiments to show that there are actually reduced levels of MyD88 protein or signaling in those mice. Next, Poly(I:C) and LPS are added to the UV-inactivated virus vaccine. Not only does immunization of mice with this preparation result in high affinity neutralizing antibody, but it protects those mice from ERD following re-exposure to RSV.

So if the alum included in the RSV vaccine is not a strong enough adjuvant to induce affinity maturation, why do any vaccines work, since alum is the only adjuvant currently used in humans? Some vaccines, like polio, are live attenuated virus, so they produce their own TLR ligands. The authors argue that the difference between successful and unsuccessful inactivated virus vaccines may be the affinity of the live virus for its cellular receptor. Influenza virus, for example, has low affinity for its receptor, so relatively low affinity antibody can be neutralizing, and indeed, influenza vaccnation doesn't induce affinity maturation. RSV, on the other hand requires a higher affinity antibody for neutralization. Whether or not virus-receptor affinity actually explains vaccine outcome, this paper serves as a nice proof-of-principle for the incorporation of our current knowledge of TLR biology into vaccine development, and addresses a 40-year-old open question in the process.

References:

Delgado MF, Coviello S, Monsalvo AC, Melendi GA, Hernandez JZ, Batalle JP, Diaz L, Trento A, Chang HY, Mitzner W, Ravetch J, Melero JA, Irusta PM, Polack FP. Lack of antibody affinity maturation due to poor Toll-like receptor stimulation leads to enhanced respiratory syncytial virus disease. Nature Medicine 15, 34 - 41 (2008)

Kim, H. W., J. G. Canchola, C. D. Brandt, G. Pyles, R. M. Chanock, K. Jensen, R. H. Parrott. Respiratory syncytial virus disease in infants despite prior administration of antigenic inactivated vaccine. American Journal of Epidemiology 89, 422-434 (1969).