Technical Comments

Response to Comment on "Stress in Puberty Unmasks Latent Neuropathological Consequences of Prenatal Immune Activation in Mice"

Science  17 May 2013:
Vol. 340, Issue 6134, pp. 811
DOI: 10.1126/science.1238060


Lazic criticizes the statistical analyses used to support the conclusions in our mouse model. His theory-biased criticism is disproportionate in view of the robustness of our findings (even if different statistical methods are applied) and falls short in explaining the postpubertal onset of effects.

In his Technical Comment, Lazic raises an essential but not novel issue related to the (in)appropriateness of experimental designs and statistics in rodent developmental biology (1). Most rodents, including mice and rats, are multiparous species and normally produce litters of up to 8 to 10 offspring. Compared with pups from other litters, littermates share similar, but not identical, antenatal (e.g., in utero physiology) and postnatal (e.g., maternal physiology and behavior) environments. Therefore, littermates are to some degree interdependent, a situation that can produce pervasive and persistent litter effects. In an influential article by Zorrilla (2), which was surprisingly not cited in Lazic’s Comment, a number of solutions have been presented as to how one can best avoid false interpretations when using multiparous species in developmental biology. Some of these issues are reiterated here against the background of our recent study (3).

We developed an environmental two-hit model in mice, in which the first experimental manipulation targeted pregnant dams, whereas the second manipulation was given to the resulting offspring (3). In brief, the first environmental hit was composed of prenatal viral-like immune activation induced by maternal administration of the synthetic double-stranded RNA poly(I:C) (polyriboinosinic-polyribocytidilic acid) during mid-gestation. Pregnant control mothers received corresponding vehicle (physiological pyrogen-free saline) solution. Offspring born to poly(I:C)–treated or vehicle-treated mothers were weaned on postnatal day (PND) 21. They were then left undisturbed (i.e., nonstressed) or exposed to variable and unpredictable stress between PND 30 and 40, the latter of which corresponded to the second environmental hit (3). As correctly pointed out by Lazic, littermates were housed two to three animals per cage and were then randomly assigned on a per cage basis to a stress or no-stress control condition. We have used two to three littermates per cage to avoid the negative influences of social isolation by single caging, which are well known to affect brain and behavioral development, especially in rodents (4).

One might argue now that the effects of prenatal poly(I:C) versus control treatment are best captured when using one offspring per litter only. In this case, the number of offspring per prenatal treatment condition [poly(I:C) or vehicle] would be identical to the number of treated mothers in each experimental group, and the offspring can easily be treated as independent observations in the statistical analyses. However, given the aforementioned problems associated with social isolation by single caging, we feel that this is not an appropriate solution here. Another possibility would be to use group caging, in which one littermate is picked out for the subsequent stress procedure. It is self-explanatory, however, that the latter experimental design would drastically inflate the number of laboratory animals to be used, and consequently, it is not an ethically feasible solution.

Contrary to the five anonymous experts who have reviewed our manuscript, Lazic suggests that the number of litters and not the number of offspring is the sample size when comparing the effect of (prenatal) poly(I:C) versus control (treatment). From a biological point of view, this argument holds true only partially in the present context because littermates from one specific litter cannot simply be considered as "technical replicates." Indeed, upon maternal administration of poly(I:C) during pregnancy, each individual fetus reacts to the maternal immune challenge by turning on or shutting down its innate gene expression machinery (5, 6). This fetal response is not only dependent on the maternal manipulation but is further regulated to a great extent by the individual fetal system, including the surrounding amniotic fluid (7, 8). It is therefore not surprising that the precise fetal (immune) response to maternal poly(I:C) administration varies significantly between individual fetuses coexisting in a single uterine horn (58). Thus, the individual offspring has a biological meaning even for multiparous species, and it is indeed the individual offspring that attracts the primary scientific interest as compared to the outcomes analyzed for whole litters.

Even if one were to analyze the data according to Lazic’s theory-biased recommendations, the reported findings and conclusions drawn from our study remain unchanged. In what follows, we present an example of the effects of prenatal immune activation and peripubertal stress on sensorimotor gating in the form of prepulse inhibition (PPI). Even though significant, the reported changes in PPI after immune challenge and stress were arguably statistically weaker compared with the outcomes in many other behavioral and neuroanatomical tests. Therefore, the PPI data set represents a valid example to test whether our findings might have been compromised by false positive outcomes. Following Lazic, one should consider the number of litters (and not the number of offspring) as the sample size and pool the cages. This statistical exercise yields a significant interaction between prenatal immune activation and stress [F(1,25) = 5.309, P = 0.032, N = 7 to 8] on percent PPI in adulthood, consistent with our original analysis using the number of offspring as the sample size [F(1,65) = 4.233, P = 0.044, N = 16 to 19] [see figure 1C and table S8 in (3)]. The equivalence of these statistical outcomes is not surprising given that our study included large numbers of offspring originating from an appreciable number of multiple independent litters.

We would also like to emphasize that the experimental manipulations of interest led to significant changes in several behavioral functions only when the offspring reached the adult stage of development, whether or not the data are analyzed using Lazic’s suggestions. Indeed, the same manipulations did not induce multiple behavioral abnormalities in pubescence despite the fact that we used (i) an identical experimental design, (ii) identical statistical methods, and (iii) comparable group and litter sizes. To our surprise, a critical evaluation of these age-dependent effects is missing in Lazic’s Comment, perhaps because his criticism falls short in explaining the post-pubertal onset of the behavioral changes.

In conclusion, we feel that Lazic’s theory-biased criticism is disproportionate in view of the robustness of our findings and falls short in explaining the age-dependent manifestation of effects.


View Abstract


Navigate This Article