To test the generality of our findings, we submitted three other RM studies with different materials and protocols to the same analysis. Source memory correlated positively with pHPC volume in both hemispheres and negatively with aHPC volume in the left hemisphere (Figure 1). To distinguish between absolute hippocampus volume and the relative size of pHPC, we calculated HPC (pHPC + aHPC) and pHPC (pHPC/aHPC) volume ratio in both hemispheres. Source memory correlated positively with pHPC volume ratios in both hemispheres, but not with HPC (Figure 1). To validate this result and probe for possible dissociations between recollection and

nonrecollection forms of memory, we obtained three published data sets with RM measures and anatomical brain images.

The behavioral measures (see Table S2 for descriptive information) were source and recognition memory for scenes (Poppenk et al., 2010b), remember/know recognition for pairs Olaparib molecular weight of words (Cohn et al., 2009), and remember/know recognition for picture-word pairs (Skinner et al., 2010). We derived RM (R) and familiarity (d’) estimates for each study: right pHPC volume ratios, but not HPC, predicted R in all three data sets ( Figure 1F). Effects of d’ resembled weaker and less reliable R effects among hippocampal measures ( Figure 1G; cf. entorhinal cortex in Figure S1). Correlations between aHPC and RM were consistently negative, CB-839 ic50 but not reliable, within single data sets. For a more sensitive test, we conducted a multilevel analysis that combined data points from all four studies (Supplemental Experimental Procedures). We found that the negative correlation was reliable in both hemispheres (Figures 2A and S2A). Greater pHPC volumes and volume ratios predicted better RM, whereas HPC did not. Also, two of the data sets included a WAIS-III digit span test. In line with the view that the hippocampus is less pivotal for working memory than RM, no correlations with digit span were observed (Figures 2B and S2B). These results confirmed our hypothesis that pHPC volume, but not HPC, would predict RM and suggested that this prediction

was selective to long-term memory. Because several of our hippocampal measures predicted RM, we explored whether their Thymidine kinase predictive power could be combined to form a stronger model. However, a forward stepwise version of our multistudy analysis (Supplemental Experimental Procedures) included only the right pHPC volume ratio in the final model, suggesting that our metrics shared most of their predictive power. In light of this shared variance, it is interesting to consider that the correlation we observed between pHPC volume ratios and RM may have been driven by the relative position of the uncal apex along the longitudinal hippocampal axis because longer segments of the hippocampus could be expected to include greater proportions of hippocampal volume.