Plasma concentrations of anti-inflammatory cytokine TGF-β are associated with hippocampal structure related to explicit memory performance in older adults

Human cognitive abilities, and particularly hippocampus-dependent memory performance typically decline with increasing age. Immunosenescence, the age-related disintegration of the immune system, is increasingly coming into the focus of research as a considerable factor contributing to cognitive decline. In the present study, we investigated potential associations between plasma levels of pro- and anti-inflammatory cytokines and learning and memory performance as well as hippocampal anatomy in young and older adults. Plasma concentrations of the inflammation marker CRP as well as the pro-inflammatory cytokines IL-6 and TNF-α and the anti-inflammatory cytokine TGF-β1 were measured in 142 healthy adults (57 young, 24.47 ± 4.48 years; 85 older, 63.66 ± 7.32 years) who performed tests of explicit memory (Verbal Learning and Memory Test, VLMT; Wechsler Memory Scale, Logical Memory, WMS) with an additional delayed recall test after 24 h. Hippocampal volumetry and hippocampal subfield segmentation were performed using FreeSurfer, based on T1-weighted and high-resolution T2-weighted MR images. When investigating the relationship between memory performance, hippocampal structure, and plasma cytokine levels, we found that TGF-β1 concentrations were positively correlated with the volumes of the hippocampal CA4–dentate gyrus region in older adults. These volumes were in turn positively associated with better performance in the WMS, particularly in the delayed memory test. Our results support the notion that endogenous anti-inflammatory mechanisms may act as protective factors in neurocognitive aging.


Introduction
Structural and functional alterations of the hippocampusdependent memory system in aging are a well-replicated finding (Gorbach et al. 2017;Nyberg 2017), and there is mounting evidence for age-related immune dysregulation as a potential factor contributing to age-related cognitive decline and ultimately clinically relevant memory disorders (Brosseron et al. 2022). Normal aging, while not usually considered a disease (Rattan 2014), is accompanied by a low-grade pro-inflammatory state commonly termed "inflammaging" (Franceschi et al. 2000). Life-long exposure to an accumulating antigen load is thought to act as a chronic stressor on the immune system and to promote chronic, asymptomatic, and inflammatory activity (De Martinis et al. 2005). Aging cells frequently enter a state of senescence, which is characterized by a silenced cell cycle and a pro-inflammatory phenotype ("senescenceassociated secretory phenotype", SASP). SASP is defined by senescence-induced increased secretion of pro-inflammatory molecules like cytokines or chemokines (Basisty et al. 2020) (see http:// www. saspa tlas. com/), which in turn induce senescence in neighboring cells (Nelson et al. 2012;Acosta et al. 2013). Furthermore, reduced adaptive immunity in old age is thought to further stimulate activity of the innate immune system as a compensatory mechanism (Fülöp et al. 2017). The resulting pro-inflammatory state increases the risk for age-related diseases like cancer or cardiovascular disease and for cognitive dysfunction (Tracy 2003;Akbaraly et al. 2013). Notably, some individuals show relatively preserved memory performance in old age (Nyberg and Pudas 2019), and research in overcentenarians suggests that balanced anti-inflammatory activity in addition to the age-typical pro-inflammatory phenotype may constitute a protective factor counteracting the negative effects of inflammaging (Franceschi 2007). A prominent anti-inflammatory cytokine is the transforming growth factor β (TGF-β). Mammalian TGF-β protein exists in three different isoforms (TGF-β 1 , TGF-β 2 , and TGF-β 3 ), with TGF-β 1 being most abundant (Dobaczewski et al. 2011). TGF-β 1 exerts considerable anti-inflammatory and neuroprotective effects (Preller et al. 2007;Harder et al. 2014), and there is evidence that TGF-β 1 is further involved in plasticity processes related to learning and memory. For example, mice treated with TGF-β 1 before being exposed to amyloid-β exhibit less synaptic loss and better-preserved memory (Diniz et al. 2017), and, in rats, TGF-β 1 protects against amyloid-β-dependent neurodegeneration and release of pro-inflammatory cytokines (Shen et al. 2014;Chen et al. 2015). Additionally, TGF-β 1 can induce the regrowth of damaged neurons after axonal injury (Abe et al. 1996). TGF-β 1 knock-out mice express lower levels of the synaptic marker synaptophysin and the plasticity-related protein laminin as well as lower synaptic density in hippocampal neurons (Brionne et al. 2003). Inhibition of TGF-β 1 signaling pathways in the mouse hippocampus leads to reduced expression of long-term potentiation (LTP) and impaired object recognition (Caraci et al. 2015), and further studies suggest that TGF-β 1 is involved in the expression of the late phase of LTP (Caraci et al. 2015;Mikheeva et al. 2019;Nenov et al. 2019). Despite considerable evidence for a contribution of TGF-β 1 to cellular mechanisms of learning and memory in rodents, little is known about a potential role of TGF-β 1 in human neural plasticity and memory. There is, however, evidence for a protective role of TGF-β 1 in neurodegenerative and neuroinflammatory diseases, particularly in Alzheimer's disease and multiple sclerosis (Martínez-Canabal 2015; Diniz et al. 2017).
Based on the well-documented role of TGF-β 1 in rodent hippocampal plasticity and memory and its neuroprotective effects in human disease, we aimed to assess a potential relationship between TGF-β 1 concentrations, hippocampal structure and learning and memory performance in healthy humans, with a focus on older adults. To control for concurrent pro-inflammatory activity, we additionally determined plasma levels of the inflammation marker C-reactive protein (CRP) and of the pro-inflammatory cytokines interleukin 6 (IL-6) and tumor necrosis factor α (TNF-α). As cognitive measures, we employed well-established neuropsychological tests of explicit memory (Verbal Learning and Memory Test, VLMT; Wechsler Memory Scale, WMS) and performed automated hippocampal segmentation and volumetry of hippocampal subfields (Quattrini et al. 2020) using magnetic resonance imaging (MRI) in a previously described cohort of young and older adults (Soch et al. 2021a;b). We hypothesized that higher TGF-β 1 plasma concentrations would be associated with larger hippocampal volumes and better memory performance, particularly in older adults. Since hippocampal neurogenesis occurs primarily in the subgranular zone (SGZ) of the dentate gyrus (DG) and the functionally related CA4 region (Bond et al. 2021), we additionally conducted a focused analysis in those subregions. According to previous studies showing that the volumes of the input regions of the hippocampus (i.e., DG, CA4, and CA3 regions) are positively correlated with verbal memory (Mueller et al. 2011;Travis et al. 2014;Aslaksen et al. 2018;Kirchner et al. 2023), we further hypothesized that larger volumes of the CA4 region and DG would correlate with better memory performance in the VLMT and WMS.

Recruitment
Participants were recruited via local press in Magdeburg, the online presence (www. lin-magde burg. de) and social media of the Leibniz Institute for Neurobiology, as well as flyers in shopping centers and at Otto von Guericke University events. Individuals interested in participating provided information about potential contraindications for participation via telephone interview before they were invited to the study. All procedures were approved by the Ethics Committee of the Medical Faculty of the Otto von Guericke University Magdeburg and conducted according to their guidelines (ethics approval number 33/15). All subjects provided written informed consent to participate in accordance with the Declaration of Helsinki (World Medical Association Declaration of Helsinki: ethical principles for medical research involving human subjects 2013).

Participants
In the present study, we investigated participants from a previously described cohort consisting of neurologically and psychiatrically healthy young and older adults (Table 1) who underwent multimodal phenotyping, including neuropsychology, MRI, and investigation of blood-based biomarkers as part 1 3 of the Autonomy in Old Age research alliance. A detailed characterization of the cohort and description of testing procedures has been reported previously (Soch et al. 2021a;b;Richter et al. 2023). All participants were right-handed according to self-report. The Mini-International Neuropsychiatric Interview (M.I.N.I.; Sheehan et al. 1998; German version by Ackenheil et al. 1999) was used to exclude present or past psychiatric disorders. Further contraindications for participation included alcohol or drug abuse, the use of neurological or psychiatric medication, and major psychosis (schizophrenia, bipolar disorder) in a first-degree relative. For the purpose of the present study, additional contraindications were chronic infectious, autoimmune or other inflammatory diseases (e.g., Crohn's disease, rheumatic diseases, Hashimoto's thyroiditis, or celiac disease), as well as regular use of immunomodulatory or antiinflammatory medication. After excluding participants with missing data, a total of 142 participants (57 young, 85 older) were available for data analysis.

General procedure of testing
After signing consent and privacy forms, participants completed questionnaires on general health and MRI contraindications, followed by the M.I.N.I. (see above). Participants older than 50 years additionally underwent the Mini-Mental State Examination (MMSE) (Folstein et al. 1975) to rule out MCI or dementia. Afterward, all participants completed the Multivocabulary Intelligence Test (MWT-B) (Lehrl 2005). This was followed by the actual computer-based cognitive test battery, which included a German version of the VLMT (see 2.2.2) (Helmstaedter 2001) and a German auditory version of the WMS for logical memory (see 2.2.3) (Härting et al. 2000). For a comprehensive description of the test battery, see (Richter et al. 2023).

Verbal Learning and Memory Test (VLMT)
The VLMT includes two lists of 15 semantically unrelated words, a study list, and a distracter list (Helmstaedter 2001). The experiment was divided into a learning phase and a recall phase. In the learning phase, the words of the first list were presented consecutively visually. After the presentation of all words, participants were asked to write down each word they could remember. This procedure was repeated five times in succession. Next, a second list was presented once, followed by a written recall. This list served as a distractor list and was followed by the recall phase, in which the words from the first list were to be written down. Further recall phases followed after 30 min and after 24 h.

Wechsler Memory Scale (WMS): logical memory
The subscale Logical Memory of the WMS was implemented as a slightly modified, auditory version (Härting et al. 2000). The test persons listened to two short stories over headphones, which they were asked to write down immediately after listening. Recall tests took place 30 min and 24 h later. The recalled stories were rated by two independent experimenters according to an evaluation sheet with 25 items (i.e., details from the stories), and thus, a maximum of 25 points could be obtained per story and recall delay.

Sample collection, processing, and storage
Venous blood samples were collected from all participants by healthcare professionals after informed consent, including two plasma tubes with sodium citrate as anticoagulant. Platelet-poor citrate plasma was prepared using a standardized two-step separation method (Reinhold et al. 1997). Citrate plasma samples were stored as aliquots in Eppendorf tubes at -80 °C until measurement. Whenever the interval between the two test days was 7 days or more, a blood sample was taken for both test days.

Enzyme-linked immunosorbent assay (ELISA)
Plasma concentrations of TGF-β 1 were determined via enzyme-linked immunosorbent assay (ELISA). Additional ELISA measurements were carried out to determine plasma levels of CRP and the pro-inflammatory cytokines IL-6 and TNF-α. Commercially available Quantikine ® ELISA kits (R&D Systems Inc., Minneapolis, MN, USA) were used for this purpose. For IL-6 and TNF-α, the high-sensitivity variant was used, as low plasma concentrations of these cytokines were expected in healthy individuals, and the high-sensitivity variant is characterized by a lower minimum detection dose.

Volumetric measurement of hippocampal subfields
Automated volumetric analysis of the individual participants' hippocampi and their subfields was performed using FreeSurfer 6.0 (Fischl 2012) and the module for segmentation of hippocampal subfields, which is based on both in vivo scans of human participants and ex vivo scans of hippocampi from autopsy specimens (Iglesias et al. 2015). Previous analyses have confirmed the robustness of this protocol across different MRI scanners (Quattrini et al. 2020).
In addition to the T1-weighted MPRAGE images, high-resolution coronal T2-weighted images (see 2.4.1) were used to improve segmentation accuracy (Dounavi et al. 2020).
To investigate a potential direct association of pro-and anti-inflammatory cytokines with learning and memory performance in the VLMT and in the WMS Logical Memory subscale, multiple linear regressions were computed with the concentrations of TGF-β 1 , CRP, IL-6, and TNF-α as well as age and gender as independent variables. An additional independent variable was the infection history of the subjects (i.e., self-report of "cold, flu, urinary tract infection or similar" or vaccination) within 4 weeks prior to testing (hence referred to as "immune event history"). The dependent variables were learning and memory performance and delayed recall in the VLMT and the WMS. The measure of learning performance for the VLMT was the sum of remembered items from learning sessions 1-5. For the WMS, the sum of the remembered items from both stories in the learning session served as measure of learning performance. Delayed recall performance was quantified as the number of recalled words or story items in the 24-h recall of the VLMT and WMS, respectively.
Statistical analysis of the relationship between TGF-β plasma concentrations and MRI data (i.e., hippocampal volumes and individual hippocampal subfields) was performed analogously to the behavioral data analysis. Multiple linear regressions were calculated with the concentrations of TGFβ, IL-6, TNF-α, and CRP as well as age, sex, and immune event history as independent variables and the volumes of the hippocampus or hippocampal subregions (DG, CA) as dependent variables, separately for each hemisphere. A correction for the false discovery rate (FDR) was applied for the number of regression analyses (N = 6). To assess for a potential association between hippocampal structure and memory performance, Pearson's correlations were computed between the volumes of subregions that showed a robust association with cytokine plasma concentrations (p < 0.05, FDR-corrected) and performance in the VLMT (learning and delayed recall) and in the WMS (immediate and delayed recall).

Results
Age and gender differences in cytokine levels, hippocampal volumes, and memory performance Table 1 displays the demographic data, cytokine levels, hippocampal volumes, and performance in the VLMT and WMS, separately for the two age groups. A MANOVA with age and gender as fixed factors and TGF-β 1 , IL-6, TNF-α, and CRP as dependent variables revealed significant effects of age group (Wilks' Λ = 0.789; F 4,135 = 9.00; p < 0.001) and gender (Wilks' Λ = 0.919; F 4,135 = 2.97; p = 0.022) as well as an age-by-gender interaction (Wilks' Λ = 0.901; F 4,135 = 3.67; p = 0.007). Post hoc two-sample t tests revealed significantly higher levels of IL-6, and TNF-α in older compared to young adults, but no age differences in TGF-β 1 , or CRP plasma concentrations (Table 1). Significant genderrelated differences were observed for CRP in the younger cohort only, with higher levels in women (young: t 37.6 = 4.17; p < 0.001; older: t 83 = 0.46; p = 0.636). No gender differences in either age group were found for TGF-β 1 , IL-6, or TNF-α plasma concentrations (all p > 0.061).
Volumes of the hippocampus and of CA4 and DG subregions showed highly significant differences as a function of both age group (Wilks' Λ = 0.618; F 6,133 = 13.70; p < 0.001) and gender (Wilks' Λ = 0.880; F 6,133 = 3.04; p = 0.008) as well as a significant age-by-gender interaction (Wilks' Λ = 0.905; F 6,133 = 2.33; p = 0.036). Table 1 displays the results of the post hoc t tests, which showed that older adults had significantly lower volumes of the whole hippocampus bilaterally as well as of all subregions tested except for the right CA4 region. Regarding gender effects, post hoc t tests in older adults revealed significantly larger volumes of the whole hippocampus and of the subfields in men compared to women (all p < 0.043). In young adults, the same tendency was observed, but largely only at a trend level (0.019 < p (one-tailed) < 0.056). This is in line with the previous studies showing larger hippocampal volumes in males, which can be attributed to larger intracranial or total brain volumes (Tan et al. 2016).
As expected, memory performance was lower in older compared to young adults. A MANOVA with VLMT learning performance and 24 h delayed recall performance as dependent variables revealed a highly significant effect of age group (Wilks' Λ = 0.459; F 2,137 = 80.74; p < 0.001) but no effect of gender and no interaction (all p > 0.120). Post hoc two-sample t tests showed significantly higher performance in young compared to older adults in both learning and 24 h delayed recall (Table 1). Similarly, WMS logical memory performance was significantly lower in older compared to young adults (main effect of age group: Wilks' Λ = 0.774; F 2,137 = 20.02; p < 0.001; see Table 1 for post hoc two-sample t tests), but no effect of gender and no interaction (all p > 0.307).

Association of TGF-β plasma concentrations with hippocampal structure
As TGF-β has previously been described to affect hippocampal neurogenesis, which occurs primarily in the SGZ, we hypothesized the subregions of the hippocampus most likely to be related to TGF-β 1 plasma concentrations would be the DG and the CA4 region. To investigate potential associations between plasma concentrations of TGF-β 1 or other cytokines (IL-6 and TNF-α) and hippocampal structures, we computed multiple linear regressions with the volumes of the entire hippocampus and the DG and CA4 regions as the dependent variable, separately for each hemisphere. Immune event history, age, and gender were included as additional independent variables. TGF-β 1 plasma concentrations were significantly positively correlated with the volumes of the CA4 and DG subregions as well as the volumes of the entire hippocampus in both hemispheres (all p < 0.050, FDR-corrected) in older adults. Other factors influencing the volumes of these regions in older adults were age and gender (all p < 0.032). Table 2 displays the adjusted regression coefficients and significance levels for all immune markers and anatomical regions, and Fig. 1 depicts the correlations between TGF-β 1 plasma concentrations and hippocampal volumes. However, no effects of TGF-β1 plasma concentrations, age, or gender could be observed in young adults (all p > 0.060, uncorrected), suggesting that the positive relationship between TGF-β 1 plasma levels and the hippocampal structure was restricted to older adults.

Correlations of hippocampal subfield volumes with memory performance
To assess a potential relationship between the volumes of the hippocampus with memory performance in regions associated with immune markers, Pearson's correlations between volumes and memory performance in the VLMT and WMS were computed for all regions showing a robust relationship (FDR-corrected p < 0.05) with at least one immune marker. As no robust associations were found in young adults, these analyses were restricted to the older participants in our cohort. We observed a positive correlation of the WMS Logical Memory scale with the volume of the right DG (p = 0.025, two-tailed) and, as a trend, also of left DG and the CA4 region bilaterally as well as the volume of the whole right hippocampus (all p < 0.05, one-tailed; see Fig. 2). Moreover, significant positive correlations with memory performance in 24-h delayed recall test of the WMS were observed for the volumes of the dentate gyrus bilaterally and the right CA4 region (all p < 0.05, twotailed) and, as a trend, also for the left CA4 region (p = 0.042, two-tailed). On the other hand, for the VLMT, no significant correlations with the volumes of any of the hippocampal regions investigated were detected (all p > 0.096).

TGF-β plasma concentrations and memory performance
In both the learning and the 24 h delayed recall phases of the VLMT, regression analysis confirmed worse performance of older adults compared to young adults (linear regression, complete sample, effect of age; learning: stand-   Fig. 1 Correlation of TGF-β 1 plasma concentrations with hippocampal subfield volumes. Pearson's correlations are shown for the dentate gyrus (DG), CA4, and whole hippocampus, separated by hemisphere. *p < 0.05, two-tailed; **p < 0.01, two-tailed CI = [-0.60 to -0.17], p < 0.001; delayed recall: standardized β = -0.28, CI = [-0.51 to -0.06], p = 0.014). As with the VLMT, however, no effect of TGF-β 1 plasma concentrations on memory performance was found (immediate recall: standardized β = -0.00, CI = [-0.35 to 0.06], p = 0.157; delayed recall: standardized β = 0.00, CI = [-0.30 to 0.10], p = 0.328). There were also no significant effects of plasma levels of other cytokines or CRP or of immune event history on either immediate or delayed recall performance in the WMS, nor an effect of gender (all p > 0.165).

Discussion
Our results suggest that plasma concentrations of the antiinflammatory cytokine TGF-β 1 are associated with hippocampal CA4 and dentate gyrus volumes in older adults. The volumes of these hippocampal regions were in turn associated with immediate and particularly delayed recall performance in the WMS Logical Memory scale in the older participants.

A potentially protective role for TGF-β 1 in the aging hippocampus
While young and older participants did not per se show agerelated differences in TGF-β 1 plasma concentrations (see Table 1), specifically in older adults, higher concentrations of TGF-β 1 were associated with hippocampal structure and particularly the volumes of the hippocampal CA4 and DG subregions. Previous rodent studies and investigations in human neurodevelopmental disorders have pointed to a role for TGF-β signaling in hippocampal development and maintenance (Stegeman et al. 2013;Oishi et al. 2016;Johnson et al. 2020). In adult mice, TGF-β can promote hippocampal LTP, particularly the transition from early to late-phase  (Caraci et al. 2015;Nenov et al. 2019), as well as adult neurogenesis in the DG (Gradari et al. 2021). Given the agedependence of the association between TGF-β 1 plasma concentrations and hippocampal structure in the present study, one possibility could be that effects of impaired TGF-β signaling during development may be compensated for during young adulthood, but, during aging, the capacity for compensation decreases. On the other hand, it must be noted that TGF-β 1 plasma concentrations were measured at or around the time of testing in our study, and we cannot make statements about long-term availability of TGF-β 1 . Therefore, in our view, a more likely or perhaps additional explanation for the age-dependence of the observed association could be related to the prominent role of TGFβ 1 signaling in CNS, but also systemic inflammation. It has been recognized for over 2 decades that microglia and astrocytes in particular play a central role as mediators between peripheral, low-grade inflammation and the cellular mechanisms of learning and memory (Yirmiya and Goshen 2011). They also express increased pro-and antiinflammatory cytokines in aging and respond to peripheral inflammatory processes (Sierra et al. 2007). For example, microglia secrete the anti-inflammatory cytokine IL-10 in response to an inflammatory stimulus, which in turn induces astrocytes to produce TGF-β, resulting in an inhibition of the microglial inflammatory response (Norden et al. 2014). However, aging astrocytes frequently show a blunted response to IL-10, resulting in lower levels of TGF-β 1 expression (Norden et al. 2016) and, consequently to a shift toward a pro-inflammatory state in microglia (Diniz et al. 2017), which has a negative impact on neural plasticity (Kempermann and Neumann 2003;Yirmiya and Goshen 2011). While this could in principle explain a relationship between higher TGF-β 1 plasma levels and larger volumes of the hippocampal CA4 region and DG, an explanation primarily based on neurogenesis may be questionable, as recent investigations cast doubt on the robustness of findings on adult neurogenesis in humans (Sorrells et al. 2021). Nevertheless, the role of chronic inflammation in synaptic dysfunction and loss and ultimately neurodegeneration is a well-documented finding (Yirmiya and Goshen 2011;Daulatzai 2014), and increased TGF-β 1 levels have been found in studies of pharmacological neuroprotection (Hosseini et al. 2018;Wiciński et al. 2018). Furthermore, TGF-β 1 also promotes the outgrowth of dendrites (Battista et al. 2006;Mathieu et al. 2010;Luo et al. 2016).
One limitation of the present study is that we could measure TGF-β 1 concentrations only in plasma but not, for example, in cerebrospinal fluid (CSF). There is surprisingly little available literature on the relationship between CSF and plasma concentrations of TGF-β 1 . Previous studies that included measures from both CSF and blood have often treated those measures separately and not reported the correlations between them (Iłzecka et al. 2002;Liu et al. 2022). One study in patients with Alzheimer's disease explicitly mentioned an absence of such correlations (Motta et al. 2007), but in that study, peripheral blood levels of TGF-β 1 were measured in serum and were thus most likely largely attributable to TGF-β 1 released from degranulated platelets (Grainger et al. 2000;Iłzecka et al. 2002). In the present study, we used a protocol that avoids contamination from platelet-derived TGF-β 1 (Reinhold et al. 1997), but a potential relationship with CSF TGF-β 1 remains nevertheless elusive. Furthermore, it must be noted that TGF-β 1 is not able to penetrate the intact blood-brain barrier (BBB) (Kastin et al. 2003). On the other hand, BBB integrity declines with age, even in healthy individuals. The hippocampus and in particular the CA1 region, but also the DG seem to be particularly affected by age-related BBB permeability, while the BBB remains comparably intact in neocortical areas (Montagne et al. 2015). Furthermore, increased BBB permeability in aging elicits increased activation of TGF-β signaling pathways and increased expression of TGF-β 1 in the hippocampus (Senatorov et al. 2019). Additionally, TGFβ 1 has been implicated in angiogenesis and BBB assembly (Diniz et al. 2017), and might increase reactively in response to BBB damage.

Hippocampal subregion structural integrity and memory performance in old age
While we observed no direct effect TGF-β 1 plasma concentrations on memory performance, we found that the volumes of the hippocampal regions shown to be associated with TGF-β 1 levels correlated with immediate and particularly delayed recall performance in the WMS Logical Memory scale in the older participants. Previous volumetric segmentation of hippocampal subregions and their association with mnestic abilities have revealed that the volumes of the input regions of the hippocampus (i.e., the DG and the CA4 and CA3 regions) are positively correlated with learning performance in well-established memory tests (Mueller et al. 2011;Travis et al. 2014;Aslaksen et al. 2018). On the other hand, the volumes of hippocampal output regions (i.e., CA1 and subiculum) are more related to delayed memory retrieval (Mueller et al. 2011;Aslaksen et al. 2018). This distinction is further supported by a functional MRI (fMRI) study on memoryrelated activations of hippocampal subregions (Eldridge et al. 2005). Nevertheless, caution is warranted, as other works have also yielded relationships between CA1 volume and learning performance as well as DG and CA4 volumes with performance in delayed retrieval (Aslaksen et al. 2018). Along the same line, a high-resolution fMRI study of successful encoding of information showed that encoding success correlated with activity in the CA1 region and subiculum, while activity in the DG, CA2, and CA3 was related to stimulus novelty (Maass et al. 2014). Evidence from rodent studies suggests that the volume of the CA4 region is associated with better spatial memory (Schwegler et al. 1990;Crusio and Schwegler 2005). To interpret these findings, one should keep in mind that "CA4" likely constitutes a misnomer, as, unlike CA1, CA3, and the small C2 region, CA4 mainly contains non-pyramidal cells and is functionally closely related to the adjacent DG (Amaral 1978;Amaral et al. 2007). As such, the region labeled "CA4" in the FreeSurfer-based segmentation of the hippocampus also contains the mossy fibers that originate from neurons of the DG, which was itself also associated with better memory performance in the older group, compatible with findings from previous studies (Zheng et al. 2018;Broadhouse et al. 2019;Kern et al. 2021;Wan et al. 2020). A larger volume in these regions might, for example, reflect both from more neurons in the DG due to better-preserved structural integrity and possibly neurogenesis as well as from a larger number of dendrites, possibly due to the role of TGF-β 1 in promoting the outgrowth of dendrites (Battista et al. 2006;Mathieu et al. 2010;Luo et al. 2016).

Clinical implications and directions for future research
Our results suggest that plasma TGF-β 1 levels are associated with a better-preserved hippocampal structure in older adults, which in turn is associated with greater episodic memory performance. Notably, our current data are based on physiological variability of TGF-β 1 plasma levels, and potential effects of raising the concentrations to supra-physiological levels, for example, by pharmacological intervention, warrant further investigation. Increased TGF-β 1 plasma levels have, for example, been found in patients with multiple sclerosis and are further increased by immunomodulatory treatment with interferon-β1b (Nicoletti et al. 1998), suggesting that the treatment might augment an already active endogenous anti-inflammatory mechanism. Little is known, however, about potential cognitive effects of elevated TGF-β 1 levels. Notably, CSF TGF-β 1 concentrations are increased in Alzheimer's disease (Swardfager et al. 2010), but their relationship with cognitive performance depends on the disease stage, with both unaffected and severely affected individuals showing a positive correlation with cognitive performance (i.e., MMSE), whereas mildly-to-moderately affected patients exhibit a negative relationship (Motta et al. 2007). Future research should further elucidate the relationship between TGF-β 1 levels and brain, particularly hippocampal, integrity in clinical populations to clarify, for example, its role in inflammatory processes related to neurodegeneration in Alzheimer's disease (Brosseron et al. 2022).
Beyond TGF-β, our results provide further evidence for a relationship between peripheral markers of inflammation and brain structure, particularly hippocampal structure, as well as neurocognitive functioning (Marsland et al. 2008;Yirmiya and Goshen 2011;Baune et al. 2012;Brosseron et al. 2023). Future studies should expand those works, for example, by employing functional MRI and derived biomarkers (Soch et al. 2021a;Richter et al. 2023). Another important direction for future research concerns gender differences in inflammaging. In the present study, we observed higher CRP levels in young, but not older, women compared to men, which have been reported previously by other groups and partly attributed to gender differences in body fat or estrogen levels (Khera et al. 2005(Khera et al. , 2009Lakoski et al. 2006). As we found no gender differences in older adults, future studies should explore the possibility that, for example, hormonal differences and their lifespan-associated changes might contribute to immunosenescence and associated neurocognitive alterations.

Conclusions
Our results suggest that TGF-β 1 plasma levels are associated with better structural preservation of the hippocampus in older adults, which is predictive for better episodic memory performance. The present data highlight the role of anti-inflammatory mechanisms as potential protective factors in neurocognitive aging. provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.