Advanced paternal age effect on offspring’s 1 reading ability: The mediating role of 2 thalamic maturation

While advanced paternal age (APA) has repeatedly been associated with higher risk 10 for neuropsychiatric disorders, its effects on cognitive processes such as reading 11 have received minimal attention. Therefore, we examined the relationship between 12 APA, offspring’s reading abilities, and brain measures in a longitudinal 13 neuroimaging study following 51 children from kindergarten through third grade. 14 APA significantly predicted reduced reading performance, independent of parental 15 reading history, socioeconomic status, home literacy environment, and birth order. 16 This effect was mediated by gray matter volume change in the left posterior 17 thalamus, predominantly the pulvinar nuclei. Complementary analyses using 18 diffusion imaging data, Neurosynth, and 1000 Functional Connectome data 19 indicated the APA-related cluster links to the dorsal attention network. These 20 findings provide novel insights into the neurocognitive mechanisms underlying APA 21 effect on reading during its earliest phase of reading acquisition and suggest future 22 avenues of research on APA-related factors, such as de novo mutation, in reading.

relationship between APA and regional brain morphometry. No significant 164 correlations were observed (all p's > 0.1). Further, whole brain analyses of regional 165 gray matter volume (GMV) at either time-point showed no significant clusters at p 166 < 0.05 corrected for Family Wise Error (FWE). Lastly, we examined the APA effect 167 Specifically, greater paternal age was associated with less GMV decrease in this 172 cohort ( Figure 2B). To verify that this APA effect was not due to confounding 173 variables, hierarchical multiple regression analyses were performed. In the first 174 model, after regressing out nuisance variables commonly controlled in longitudinal 175 reading outcome (95% confidential interval was [-0.303 -0.022] when not controlling 208 for age at t2, handedness, average of t1 and t2 pIQ, and t1RAN, and changed to [-209 0.249 -0.001] when these covariates were controlled; Figure S1). possibly because the APA-cluster also contained white matter. As presented in 218 Figure 4A, within the overlapping region, 380 voxels (69.5%) were in the 219 subdivision labeled as pulvinar nuclei, especially the medial portion, which is 220 known to have widespread connections with the inferior parietal lobule (Arcaro, 221 point to the attention network, in particular the DAN, to be the candidate brain 250 functional system associated with the APA-cluster in the left thalamus. 251 In the final step, we used diffusion imaging data available in a sub-group of 252 23 participants to determine which was more likely the candidate. Using 253 deterministic tractography, we reconstructed white matter fibers through the APA-254 cluster, covering inferior fronto-occipital fasciculus, corticospinal tract, forceps 255 major, superior corona radiata, as well as anterior and posterior limbs of the 256 internal capsule. Figure 6A shows reconstructed fibers in a representative child and 257 Figure 6B shows intersection across participants, for demonstrative purposes. More 258 importantly, the APA-cluster showed significantly stronger connectivity (defined by 259 dividing the total number of streamlines by the size of the target network) with 260 DAN than with VAN (t = 6.61, p < 0.001; Figure 6C). Finally, consistent with the 261 aforementioned results, correlation analyses showed significant positive correlation 262 between APA-DAN streamlines and PatAGE (r = 0.49, p = 0.018; Figure 6D), and 263 significant negative correlation between APA-DAN streamlines and t2READ (r = -264 0.45, p = 0.030; Figure 6E). No significant correlations were found between APA-265 VAN streamlines and PatAGE or t2READ (both p's > 0.1). 266

267
The present study provides evidence that APA is negatively associated with 268 offspring's reading by rigorously controlling for a number of potential confounding 269 factors and for the first time investigates neurocognitive mechanisms underlying 270 . CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted May 22, 2020. ; https://doi.org/10.1101/2020.05.20.105759 doi: bioRxiv preprint the APA effect on reading by using multiple neuroimaging modalities and online 271 databases/atlases. We showed that the APA effect on reading skills was 272 independent of familial factors such as parental reading history, SES (an aggregate 273 measure of family income, parental educational level and occupation), and home 274 literacy environment. We also investigated the neurobiological correlates of APA, 275 identifying it to be the grey matter development in the left thalamus. The converging evidence that APA may be a risk factor that negatively impacts reading, 287 independent of phonological processing, through an altered maturational process of 288 the left thalamus. 289

Potential mechanisms of APA effect on offspring's reading 290
In this study, we demonstrated a negative APA effect on offspring's reading abilities. 291 This finding is consistent with a study in boys with dyslexia (Jayasekara & Street,292 . CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted May 22, 2020. ; https://doi.org/10.1101/2020.05.20.105759 doi: bioRxiv preprint occupation, and family income). Furthermore, we used a battery of measures to 315 accurately assess children's reading abilities after three years of formal reading 316 instruction. We also acquired cognitive-linguistic precursors of reading (i.e., PA, 317 RAN, letter knowledge) of these children at the beginning of formal reading 318 instruction in kindergarten, allowing us to examine reading development 319 longitudinally. Our results first provide a unique picture of the negative APA effect 320 on reading, after controlling for possible confounds. Moreover, although there is a 321 possibility that fathers who are poorer reader may have children later because for 322 example, they may take time to be financially independent and hence likely have 323 poor reading children, it is not the case in the current study since we did not find a 324 significant correlation between PatAGE and paternal ARHQ. This pattern further 325 confirms the negative influence of APA on reading. Relevant to these findings, we 326 did not observe associations between paternal age and cognitive-linguistic skills (i.e., novo genetic mutations and epigenetic modifications (e.g., DNA methylation and 337 repressive histone modification) in paternal gametes (Girard et  propose that de novo mutations may be one mediator of APA effects on reading, and 372 a potential risk factor for poor reading. Future research analyzing deoxyribonucleic 373 acid (DNA) from triads exploring the relationship between de novo mutations and 374 offspring's reading phenotypes is warranted. Of relevance, a recent study used 375 whole-genome sequencing in parent-child trios and discovered de novo mutations 376 that disrupted specific genes (e.g., CHD3, SETD1A, WDR5) in individuals with 377 childhood apraxia of speech, another common neurodevelopmental disorder (Eising 378 et al., 2018). Similar studies may be a promising approach to reading research. 379 . CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted May 22, 2020. The current study demonstrated a negative APA effect on offspring's reading after 382 controlling potential confounding factors. It further revealed that this effect was 383 mediated by morphometric maturation of the left posterior thalamus, providing a 384 potential neural explanation at the macroscopic level. 385 The thalamus is an important relay center in the human brain, connecting 386 cortical and subcortical areas, receiving information from sensory cortices and 387 relaying it to higher-level association areas. Studies in normal populations with 388 cross-sectional designs have produced a mixed picture of the normal developmental 389 trajectory of the thalamus: while the relative gross volume of the thalamus 390 (normalized by brain size) was found to decrease from 4 to 18 years of age by 391 was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made  was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted May 22, 2020. ; https://doi.org/10.1101/2020.05.20.105759 doi: bioRxiv preprint examining the interaction between three major components of attention (altering, 424 orienting, and executive control) (Xuan et al., 2016). The pulvinar is also involved in 425 writing, which is related to reading and imposes a high demand of visuo-spatial 426 attention among other processes (Yuan & Brown, 2015). Consistent with this line of To date, research into the APA effect on neural networks and cognitive 444 processes is scarce. The study conducted by Shaw and colleagues focused on 445 . CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted May 22, 2020. ; https://doi.org/10.1101/2020.05.20.105759 doi: bioRxiv preprint parental age effects on cortical thickness and surface area but did not examine their 446 relationship to cognitive functions, making this study somewhat inconclusive as to 447 the role of parental age on neurocognitive processes (Shaw et al., 2012). Taking one 448 step further, our study revealed an intermediary between paternal age and a 449 specific behavioral phenotype at the neural level, offering initial insights into the 450 complex mechanisms underlying APA effects.  Additionally, this effect appeared to be independent of environmental risks such as 467 . CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted May 22, 2020. Limitations 527 In the present study, we found a negative effect of APA on offspring's reading 528 achievement. However, the results should be interpreted with cautions. First, 529 because the range of paternal age at the time of child's birth in this study was 530 restricted to 25-47 years, the findings may not necessarily be extended to children 531 with fathers on extreme ends, young or old. Of relevance, young fatherhood has also 532 been associated with adverse cognitive development of the offspring (Weiser et al., 533 . CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted May 22, 2020. ; https://doi.org/10.1101/2020.05.20.105759 doi: bioRxiv preprint 2008) but possibly due to different factors such as immature sperm and economic 534 disadvantages (Chen et al., 2008). Second, because children's reading abilities were 535 measured at grade 3, it is unknown whether the APA effect on reading will persist 536 into adulthood or is simply a developmental delay. Third, since reading performance 537 of the participants were within typical range, our results show that individual 538 differences in reading ability are associated with age of the fathers, rather than 539 direct evidence that late fatherhood is associated with RD in their offspring. Finally, 540 while we revealed the left posterior thalamus mediated the APA effect on reading, 541 we could not answer why APA (or de novo mutations) specifically impacts this 542 subcortical area. Given that the typical maturation of thalamus can be also affected 543

Conclusion 547
The current study, for the first time, examined the association between APA and 548 reading at both a behavioral and neurobiological level. We provided evidence that 549 APA is an independent factor associated with lower reading ability. We also found 550 that the APA effect on reading was mediated by maturation of the thalamus. This 551 suggests a novel neurobiological pathway for intergenerational influence on reading, 552 completing prior findings that offspring's reading is influenced by parental reading included. There was no significant difference in either familial or any behavioral 593 measures between the total cohort and any sub-groups (all p's > 0.1). The 594 investigator, approved the present study. Both informed assent and consent were 598 obtained from children and their parents/guardians. 599

Family information and behavioral measurements 600
Demographic information, family and behavioral measures are summarized in 601

Image acquisition 629
High-resolution T1-weighted images (fast spoiled gradient echo) for each child were was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made of Achievement. In each analysis, Maximum Likelihood was used as the extraction 651 method, Varimax was used as the rotation approach, and Bartlett method was used 652 to calculate factor scores. From t1 behavioral metrics, we obtained two factors using 653 the criteria of eigenvalues greater than 1 (Table S3) was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted May 22, 2020. ; https://doi.org/10.1101/2020.05.20.105759 doi: bioRxiv preprint and were named as t2READ, t2PA, and t2RAN according to the factor loading 663 scores (Table S4). 664 To test our hypothesis about the relationship between APA and reading, we 665 first performed simple correlation. Once significant correlation between PatAGE 666 and t2READ was observed, three hierarchical multiple regressions were further 667 conducted to test three hypotheses in the following order: (1) APA effect remains 668 significant after controlling for demographic variables; (2) APA effect is present 669 above and beyond other familial factors; (3) APA effect is not explained by t1 670 cognitive-linguistic skills (t1PA and t1RAN) generally known to be highly heritable 671 and is relatively independent of these precursors. Therefore, in the first model, we 672 entered demographic variables (t2 age, gender, handedness and average pIQ from t1 673 and t2) in the first step and PatAGE in the second step (Model 1 in Table 1). In the 674 second model, besides the aforementioned nuisance variables, we additionally 675  Table  679 1). In the third (final) model, t1PA and t1RAN (Hulme et al., 2005) were 680 additionally entered in the third step, just before PatAGE was entered (Model 3 in 681 Table 1), to examine whether the APA effect was present beyond t1 cognitive-682 linguistic skills. All statistics were done with SPSS 21.0 (IBM, Inc.), and p-values 683 were two-tailed while statistical significance was set at 0.05. 684 . CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted May 22, 2020. ; https://doi.org/10.1101/2020.05.20.105759 doi: bioRxiv preprint One main aim of this study was to explore the cognitive mechanisms 685 underlying the APA effect, e.g., whether the APA effect on reading was mediated by 686 cognitive-linguistic precursors such as PA. But since t1RAN and t1PA showed no 687 significant correlations with PatAGE, no further mediation models were established 688 with t1RAN or t1PA as mediators. 689

Structural image preprocessing 690
Both cross-sectional and longitudinal analyses were conducted with VBM8 691 was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted May 22, 2020. ; https://doi.org/10.1101/2020.05.20.105759 doi: bioRxiv preprint As for the longitudinal VBM analysis, 'Preprocessing of Longitudinal Data' 705 module in VBM8, which contains specific preprocessing steps was used. Intra-706 subject realignment, bias correction, segmentation, and normalization (Ashburner, 707 2007) were done sequentially as described elsewhere (Ridgway et al., 2007). After 708 applying spatial smoothing with an 8-mm full-width half-maximum Gaussian 709 kernel, we obtained maps of gray matter volume for both time-points. We generated 710 GMV maps reflecting change from t1 to t2 for further analyses (such that a positive 711 change would indicate growth from t1 to t2). 712

Whole-brain regression analyses 713
First, we examined the correlations between PatAGE and global measurements, i.e., 714 t1TIV (defined as the sum of total gray matter, white matter and cerebrospinal fluid) 715 and t2TIV. Then, we examined whether PatAGE correlated with ∆TIV between two 716 time-points (such that a positive change would indicate growth from t1 to t2) while 717 controlling for the baseline (i.e., t1TIV). After that, to examine relationships 718 between regional GMV at each time-point, as well as ∆GMV with PatAGE, voxel-719 wised whole brain regression was conducted while controlling for the effect of global 720 measurements. Specifically, t1TIV or t2TIV was controlled in cross-sectional 721 analyses for t1 and t2, respectively. In the longitudinal analysis, t1TIV and ∆TIV 722 were controlled to exclude effects from initial gross volume and its development. MatARHQ since they showed significant correlation with PatAGE. 738 Next, we examined the relationship between ∆GMV and children's t2READ 739 in the cluster that was significantly associated with PatAGE (i.e., the APA-cluster, 740 which was in the left thalamus) by using small volume correction (p-voxel < 0.005, 741 p-cluster < 0.05, topological FWE correction) while t1TIV and ∆TIV were 742 statistically controlled. The mean ∆GMV was calculated from this APA-cluster in 743 the left thalamus for subsequent ROI analyses. Then, hierarchical multiple 744 regression analyses were conducted to test for the robustness of the effect. In the 745 first model, t2READ was the dependent variable and demographic variables (t2 age, 746 gender, handedness, average pIQ), t1TIV and ∆TIV were entered in the first step. In 747 the second model, t1 cognitive-linguistic skills (i.e., t1PA and t1RAN) were further 748 . CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted May 22, 2020. ; https://doi.org/10.1101/2020.05.20.105759 doi: bioRxiv preprint entered in the second step since they were also significant predictors of t2READ in 749 previous analysis. The average ∆GMV was entered in the final step. 750

Mediation analyses 751
In the region where volumetric change significantly correlated with both PatAGE 752 and t2READ, we used mediation as the conceptually preferred model to examine 753 whether the negative impact from APA on reading was mediated by brain 754 maturation. To test indirect effects, bootstrapping (10,000 samples) was used to 755 obtain 95% confidence intervals. We first ran a basic model without controlling for was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted May 22, 2020. ; https://doi.org/10.1101/2020.05.20.105759 doi: bioRxiv preprint Next, to further understand the functional role of the APA-cluster and 791 complementary to the results from analyses using the histological and diffusion 792 imaging atlases, we examined APA-cluster-associated cortical patterns by using an 793 online database, Neurosynth (v0.5; Yarkoni et al., 2011). In particular, we 794 generated a co-activation map by including all fMRI studies in the database (N > 795 10,900) and used the whole APA-cluster as ROI. A threshold of False Discovery 796 Rate (FDR) at p < 0.01 was used to obtain significant regions that is most likely to 797 be reported in fMRI studies when the APA-cluster is also reported (i.e., forward 798 inference). In addition, we generated a seed-based whole-brain RSFC map by using  . CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted May 22, 2020. ; https://doi.org/10.1101/2020.05.20.105759 doi: bioRxiv preprint . CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted May 22, 2020. ; https://doi.org/10.1101/2020.05.20.105759 doi: bioRxiv preprint   was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made   Figure 2A). B. 1357

Figure Legends
Intersection across children with diffusion imaging data (N = 23) is shown for 1358 demonstrative purposes. In particular, only fibers (i.e., streamlines) observed in 1359 more than 25% of the subjects (i.e., 6 children) are displayed. The color bar 1360 . CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made  was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted May 22, 2020. ; https://doi.org/10.1101/2020.05.20.105759 doi: bioRxiv preprint Tables   1383  1384  Table 1 Multiple linear regression analyses examining the unique contribution of 1385 paternal age on offspring's reading performance at time-point 2 1386

Model
Step . CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted May 22, 2020. ; https://doi.org/10.1101/2020.05.20.105759 doi: bioRxiv preprint . CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted May 22, 2020. ; https://doi.org/10.1101/2020.05.20.105759 doi: bioRxiv preprint . CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted May 22, 2020. ; https://doi.org/10.1101/2020.05.20.105759 doi: bioRxiv preprint . CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted May 22, 2020. ; https://doi.org/10.1101/2020.05.20.105759 doi: bioRxiv preprint was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted May 22, 2020. ; https://doi.org/10.1101/2020.05.20.105759 doi: bioRxiv preprint Similarity with the Intrinsic Functional Networks . CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted May 22, 2020. ; https://doi.org/10.1101/2020.05.20.105759 doi: bioRxiv preprint was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted May 22, 2020. ; https://doi.org/10.1101/2020.05.20.105759 doi: bioRxiv preprint

Orthographic Processing
Liability for poor reading Population incidence . CC-BY-NC-ND 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted May 22, 2020. ; https://doi.org/10.1101/2020.05.20.105759 doi: bioRxiv preprint