This cross-sectional study supports the hypothesis that PFCs may be neuoprotective. Our data suggest that higher PFC concentrations are associated with reduced risk of cognitive limitation, especially among non-medicated diabetics.
Our current mechanistic hypothesis posits that PFCs may exert neuroprotective effects via activation of PPARγ, as PFOA and PFOS have been shown to activate human PPARα and PPARγ under some experimental conditions [3
]. Other PPARγ agonists, which include non-steroidal anti-inflammatory drugs and the anti-diabetic thiazolidinedione drugs, appear to be neuroprotective [23
], potentially through inhibition of inflammation, oxidative stress, and apoptosis, modulatation of signaling pathways involved in the pathogenesis of neurodegenerative diseases, or improvement in insulin sensitivity and reduction in blood glucose levels [4
], although it must be recognized that they may also act exclusive of PPARγ activation. In our study, we observed the strongest protective associations of PFCs on limitation due to difficulty remembering or periods of confusion within non-medicated diabetics. Conversely, we found no association in non-diabetics or diabetics using thiazolidinedione drugs for all PFCs and a moderate protective association in persons using only non-thiazolidinedione diabetes medications for three of the four PFCs. This pattern supports the idea that PFCs and thiazolidinedione drugs may share a common mechanism, PPARγ activation.
The finding that the association was concentrated in diabetics, specifically non-medicated diabetics, raises the question of whether diabetes or diabetic severity influences PFC concentrations. Hypothetically, diabetic nephropathy may impact PFC concentrations. First, if PFCs are excreted via urine, lower glomerular filtration rate could artificially raise the serum PFC concentrations. However, this would induce an adverse association given that increasing severity of diabetes would predict both higher PFC levels and worse cognition. Second, kidney dysfunction could lead to either increased water retention and higher plasma volume or dehydration due to water loss associated with loss of glucose in the urine and lower plasma volume. Higher plasma volume could artificially lower PFC concentrations in persons with more severe diabetes, inducing a protective association, while lower plasma volume could be expected to do the opposite. Third, if excess PFCs are lost because of their association with serum protein as poor kidney function leads to protein loss in urine, PFC concentrations could be lower in diabetics with greater diabetic severity, artificially inducing a protective association. However, adjustment for osmolality - a marker of plasma concentration, estimated glomerular filtration rate - a marker of kidney function, and urinary albumin – reflecting protein loss, did not appreciably change our estimates suggesting that such explanations are unlikely to account for the observed associations.
We recognize that our primary measure of cognition, self-reported limitation due to difficulty remembering or periods of confusion is not a clinical endpoint. However, in clinical settings, subjective memory complaint often motivates clinical evaluation of cognition and is generally accepted as a key feature of mild cognitive impairment [30
] and a growing body of literature has linked subjective memory complaint prospectively to poor cognition [31
], dementia [32
], neuropathology [33
], and future cognitive decline [31
]. Our measure, self-reported limitation due to poor memory or periods of confusion is a strengthened version of questions about subjective memory impairment/complaints given that it focuses on limitations
due to poor memory or confusion, rather than difficulty remembering or poor memory compared with previous performance [35
], and so is more likely to reflect current impairment. Supporting this, the prevalence of subjective memory complaints is typically ~20–60% [34
] while the prevalence of cognitive limitation in our data is 13%. In addition, variables shown in other studies to predict cognitive impairment or cognitive decline, such as age, diabetes, physical activity, and hypertension also appear to predict our outcome in multivariable models (data not shown). Further research should evaluate the association between PFCs and cognitive test scores or clinical diagnoses.
The relevant period of potential exposure susceptibility is difficult to identify, but our understanding of cognitive decline suggests a period of years to decades prior to clinical onset of dementia. More recent exposures may influence progression of brain pathology underlying cognitive difficulties, while more distant exposures may influence initiation of this pathology. The assumption that current PFC concentrations reflect past concentrations in a relevant time period for exposure is reasonable given their long half-life in humans, ranging from 2.3 years for PFOA to 8 years for PFHxS [36
The remaining limitations of our study must be noted. We did not find support for an association between PFCs and DSST scores; however, these analyses were limited by the small number of participants and the limited range of cognitive domains assessed by the DSST. As our study is cross-sectional, it is possible that poor or declining cognition impacts PFC levels, rather than the reverse, but this explanation seems unlikely given the long half-life of PFCs in humans. The potential for bias due to residual or unmeasured confounding remains. In particular, we were only able to adjust for fish consumption in sensitivity analyses; however, results from our sensitivity analyses were consistent with our main findings. We were able to control for many known or suspected confounders, including sociodemographic factors. Selection bias due to non-participation is not expected to influence our results given that people are unaware of their PFC exposures and PFCs are not known to significantly predict mortality. Misclassification of PFC exposure or cognitive impairment may contribute to the current findings, making it more difficult to detect a true effect. Finally, we were unable to disentangle the effects of individual PFCs, as all four are moderately correlated (range: 0.35 (PFNA and PFHxS) to 0.64 (PFOA and PFOS)). However, to our knowledge, this is the first paper to consider the relationship between PFCs and cognition in older adults and we believe that the difficulties discussed above are unlikely to account for a protective association.