Protein profiles in plasma: Development from infancy to 5 years of age

Little is known about the longitudinal development of different plasma protein levels during early childhood and particularly in relation to lifestyle factors. This study aimed to monitor the plasma proteome early in life and the influence of different lifestyles.


INTRODUCTION
Environment early in life has a significant impact on future health. [1] Proteins are responsible for key biological functions such as signaling and defense, in an age-related balance influenced by genes, lifestyle, and environment. [2,3] However, little is known about the longitudinal development of different plasma protein levels during early childhood, particularly in relation to environmental factors. [4] The prospective birth cohort Assessment of Lifestyle and Allergic Disease During Infancy (ALADDIN) [5] offers a unique opportunity to study development of different plasma proteins early in life as well as associations between environmental and lifestyle exposures. In 2004, we initiated this cohort by recruiting families with conventional or anthroposophic lifestyle, [5] with the purpose to identify environmental and lifestyle-related factors during pregnancy and early childhood that may constitute risk of or protection against development of different diseases, particularly allergy. Anthroposophic lifestyle is characterized by, among other things, home delivery, organic diet with live lactobacilli, restrictive use of antibiotics and vaccine, and for the infant avoidance of stressful stimuli. [5][6][7][8] The aims of this study were to (i) determine the longitudinal development of different plasma protein levels during early childhood and (ii) compare the influence of different lifestyles.

Subjects
The prospective birth cohort study ALADDIN consists of families with an anthroposophic, partly anthroposophic, or non-anthroposophic lifestyle. [5] A (1) "What kind of preschool/school will your newborn child probably go to?" (2) "Has any of the parents, no matter which type of school you have planned for your child, an anthroposophic view of life?," and (3) "Is the family's everyday life influenced by an anthroposophic view of life?" Families answering "anthroposophic school" to question 1 and "yes" to questions 2 and 3 and also attending anthroposophic antenatal clinics were defined as "anthroposophic." Families answering conventional or any other non-anthroposophic type of school to question 1, "no" to questions 2 and 3 and going to conventional antenatal clinic were defined as "non-anthroposophic." Any other combination of answers was defined as "partly anthroposophic." [5] In Data on delivery was retrieved from the Swedish Medical Birth Register, and vaccination data from medical records at child healthcare centers. [7] This study was performed in accordance with the principles expressed in Declaration of Helsinki and approved by the Regional

Collection of blood samples
Blood samples were collected venously in 4 mL sodium heparin tubes from the children at 6 and 12 months and in 9 mL sodium heparin tubes at 24 and 60 months of age. The blood samples were stored at room temperature before centrifugation for 10 min at 800 × g and the plasma was subsequently collected. The distribution of time periods at room temperature across lifestyles were similar, see Table S1. The median volumes of plasma obtained were 1.5 mL (range 0.5-2.5 mL) at both 6 and 12 months, 4.0 mL (range 0.8-4.0 mL) at 24 months, and 3.6 mL (range 2.0-4.0 mL) at 60 months. Plasma was stored at −80 • C. Blood was collected venously from all children except from seven children (10%) at 6 months of age and five children (7%) at 12 months of age where capillary blood was collected due to practical and humanity reasons. For a distribution of capillary blood samples across lifestyles, see

Protein selection and plasma profiling using antibody suspension bead arrays
For the protein profiling, a panel was designed based on in-house knowledge combined with database searches (Ensembl, Genotator, the Human Protein Atlas [HPA], Ingenuity Pathway Analysis, Intomics, and MetaCore), in general focusing on proteins expressed or detectable in blood/serum/plasma and involved in biological processes of the immune system and inflammatory responses. In total, 151 proteins were included in the panel and targeted by 243 antibodies (Table S3).
This final selection of proteins was based on antibody availability within the HPA project (https://www.proteinatlas.org). [9] One antibody per protein target was included for 78 proteins whereas 73 proteins were targeted by more than one antibody (ranging between two and five antibodies per protein target). This is a common strategy when as in this case using antibodies that were generated against different or same regions of the protein (immunogens with partial to full protein sequence coverage), for example, seen in [10][11][12]. Quality assessment of these multiple antibodies is defined in the quality control and statistical analysis section below. Most of the antibodies were affinity purified polyclonal rabbit antibodies from the HPA project, [9] with the exception of anti-CHI3L1 (polyclonal goat IgG AF2599 and monoclonal rat IgG 1 MAB25991, R&D Systems, Minneapolis, MN, USA) and anti-HPGDS (monoclonal mouse IgG 1 MAB6487, R&D Systems).
Plasma profiling was performed using a multiplexed bead-based immunoassay based on antibody suspension bead arrays. To create the bead array, antibodies were immobilized onto color-coded magnetic beads (MagPlex-C, Luminex Corp., Austin, TX, USA), as previously described. [11,13] Here, antibodies were diluted to 1.6 µg/mL in MES and coupled to beads that had been activated in N-hydroxysulfosuccinimide/1-ethyl-3-(3-dimethylaminopropyl)carbodiimide hydrochloride (Pierce, Thermo Fisher Scientific, Waltham, MA, USA) prior to the coupling. All beads were combined to create a multiplex antibody bead array and successful coupling confirmed using an R-phycoerythrin-

Quality control and statistical methods
Statistical analysis and visualization were performed in the programming language and environment for statistical computing R [14,15] and SPSS version 25 (IBM Corp., Armonk, NY, USA). The raw fluorescent intensity data were processed with the R package "MDimNormn" and function "normn_MA" to correct for potential batch effects arising from the multiple microtiter plates. [16] To assess the technical variation, the coefficients of variation (CV) were calculated for each antibody across technical replicates (four per 96-well plate, 16 in total). The median CV per 96-well plate was between 6.5% and 9.5% for the four plates, and 8.6% for the whole assay 384-well plate.
There were no plate effects, assessed by Kruskal-Wallis tests. We applied a quality control filtering step, where the data for proteins measured by multiple antibodies were assessed by Spearman rank correlation. Antibody pairs with high agreement (defined by Spearman's Rho ≥ 0.7) were included in the downstream analysis whereas pairs with low correlation (Spearman's Rho < 0.7) were excluded.
This resulted in 19 proteins supported by two-three antibodies (Table   S4). After this filtering step, a total of 97 proteins (117 antibodies) remained. A merged antibody profile was created for each of the 19 proteins by computing the mean intensity across the antibodies targeting the same protein. In the end, the data from 97 proteins were used in the succeeding statistical analysis (see Figure S1 in the Supporting Information for a functional annotation summary of these proteins).
In the univariate analysis, the non-parametric test of repeated measures, Friedman's test, was used to identify proteins with a time (age) difference, that is, which of the 97 proteins that changed significantly over time (at 6, 12, 24, and 60 months of age). First, all 70 children were compared with respect to differences between the time points in all 97 proteins. Second, stratified analyses were performed for the 97 proteins in which differences between time points were tested within each lifestyle group (the anthroposophic, partly anthroposophic, or non-anthroposophic children) separately.
To account for multiple testing, the Bonferroni method was used to adjust P-values with respect to the number of proteins tested (i.e., the two-sided P-values from Friedman's test were multiplied with 97). In the multivariate analysis, longitudinal profiles, or trends, in the data were identified with the self-organizing tree algorithm (SOTA) method using the R package "clValid." [17] Log-transformed data were based on the GO Ontology database from 2020-08-10). [18] Analysis of overrepresented GO biological process terms in the 97 measured proteins used Homo sapiens with the whole human proteome as background. The overrepresentation test of significantly altered proteins used the 97 measured proteins as background. FDR < 0.05 was considered significant.

Study population
The study population (n = 70) constituted of 15 children living in families with anthroposophic, 27 with partly anthroposophic, and 28 with non-anthroposophic lifestyle. Distribution of sex, gestational age, and birth weight were similar between the three lifestyle groups, whereas lifestyle-related exposures as motherst' diet during pregnancy, home delivery, duration of breast feeding, and time points of vaccinations differed significantly (Table 1).

Longitudinal development of plasma proteins
To study the general development of plasma proteins across the first 5 years of life irrespective of lifestyle group, we examined the relative levels of 97 proteins in all 70 children using a multiplex bead-based immunoassay in 280 samples. The levels of 68 proteins were significantly altered between any of the time points (ages of 6, 12, 24, and 60 months) (Tables S3 and S5). In the functional classification analysis of these proteins, no biological processes were statistically overrepresented, indicating that the altered proteins represented various functions as those seen for the 97 proteins ( Figure S1).
To visualize the trends over time in these significantly changing proteins, we performed SOTA clustering of the 68 proteins for further characterization ( Figure 1A). This revealed that the proteins could be grouped in four clusters, or profiles, with characteristic trends. Cluster years. See Figure 1B for examples of proteins of each cluster, and Table   S3 for information of which cluster each protein belonged to.

Plasma proteins stratified by lifestyle
We then investigated whether the lifestyle would influence the protein levels during early childhood. First, the SOTA clustering was repeated but now instead stratified by each of the three lifestyles ( Figure 2). This again revealed four clusters with similar profiles as the SOTA clustering of the non-stratified data in Figure 1, indicating that proteins developed with the same general pattern regardless of lifestyle. See Table   S3 for information of the proteins in each cluster. When comparing the proteins in each SOTA cluster across the three lifestyle groups, only few proteins were unique to a certain lifestyle ( Figure 3A). Second, to identify proteins that were significantly altered over time within each lifestyle, we performed statistical tests stratified by lifestyle (Tables S3 and S5). Here, in total 59 proteins were altered over time within one or more of the three lifestyle groups. All these 59 proteins were also altered in the group of all children, which additionally had nine more proteins that were not significant within any of the lifestyle groups ( Figure 3B). In the group of children with anthroposophic lifestyle, the levels of 30 proteins were significantly altered and 60 months), Hemophilus influenzae type b (at 3, 5, and 12 months), measles, mumps, and rubella (at 18 months and 6 years). [35] The median for the shortest period within a lifestyle group between most recent vaccine prior to blood collection was 1.2 months (at 6 months collection), 4.5 months (at 12 months collection), 3.8 months (at 24 months collection), and 15.1 months (at 60 months collection). b P-values from Fisher's exact test for categorical variables and Kruskal-Wallis test for continuous variables. c Measles, mumps, and rubella. between any of time points, whereas 56 proteins changed in the partly anthroposophic group (Tables S3 and S5). In the non-anthroposophic group, 47 proteins had altered levels (Tables S3 and S5). Out of the 59 proteins that changed in total, 28 proteins were altered indepen-dent of lifestyle group as shown in the Venn diagram ( Figure 3B and   Figure 3B and Table S6).
To further explore the 68 proteins that were altered over time,  Table S6 for further details

F I G U R E 4
Heatmap of all 70 children and the profiles of 68 proteins that were significantly altered using Friedman's tests. Hierarchical clustering on the level of samples (rows) collected at ages 6, 12, 24, and 60 months as well as on the level of proteins (columns). Samples are furthermore annotated based on the lifestyle of the family (anthroposophic = pink, partly anthroposophic = blue, non-anthroposophic = orange) and age (6-60 months corresponding to a gradient of lighter to darker green). Proteins are annotated based on the result from the SOTA clustering performed on data from all children or from each of the lifestyles separately (cluster: 1 = blue, 2 = pink, 3 = yellow, 4 = green). AU, arbitrary units (green). This indicates that the proteins belonging to SOTA cluster 1 + 2 and SOTA cluster 3 + 4 share similarities in their profiles.

DISCUSSION
This study of a unique longitudinal cohort of children with different lifestyles combined with high-throughput plasma protein profiling showed that a majority (70%) of the 97 plasma proteins analyzed were altered during infancy up to 5 years of age. The temporal changes represented four longitudinal trends of the plasma proteins during development, also following stratification of lifestyle.
There are only few studies of age-associated differences and longitudinal profiles of plasma proteins in pediatric cohorts. Age-related changes of plasma proteins were studied in neonates, infants, children, and adults in two cross-sectional studies. [2,3] In both studies, two age groups (<1 and 1-5 years) match our age interval but as the group of 1-5 years is covered by three individual time points in our data, a direct comparison is somewhat limited. Nevertheless, in Ignjatovic et al., [3] age-associated proteins were involved in processes such as the immune response, homeostasis, and hemostasis in agreement with our data. Although only proteins that were lower or higher in adults compared to the younger age groups were reported, [3] we also observed an age-effect for A2M, FGA, SERPINA1, and SERPINA3. Bjelosevic et al. [2] identified proteins involved in hematopoietic development, immune response, and physiological growth to be altered across the age groups.
Furthermore, differences between the groups of <1 and 1-5 years were identified for 27 proteins (see Table S4 in [2]). Four of these proteins were included in our analysis and three (C4A, C4B, and C9) were significantly altered over time (Table S3). What becomes evident from comparing our data to cross-sectional studies like these, [2,3] is the added resolution by including multiple time points. This is highlighted by the marked changes that we could observe between the ages of 6, 12, 24, and 60 months but that are averaged and potentially missed when combining subjects of different ages into one group.
We observed four major patterns of longitudinal protein profiles across ages of 6 months to 5 years in the proteins that were significantly different (Figures 1 and 2). In cluster 1 (Figure 1), we found multiple apolipoproteins, proteins that not only play a role in the transport and metabolism of cholesterol and lipids but also in immunity and inflammation, such as the protective role seen for APOA1. [19,20] Cluster 2 showed a decreasing trend over time ( Figure 1B) and included secreted phosphoprotein 1 (SPP1), also called osteopontin. SPP1 is involved in several processes like bone formation, wound healing, tumorigenesis, and immunological responses. [21] In agreement with our data, human plasma SPP1 was recently found to have a constant age-related decline from birth until the age of 14 years with the highest levels between 0 and 24 months of age. [22] Cluster 3, which included an increase in protein levels between 12 and 24 months, included proteins that may be induced by the immune stimulation associated with entering preschool, in Sweden usually at 1.5 years of age, as exemplified by TNF. [23] The costimulatory protein CD40, required for activating antigen-presenting cells, [24] showed a cluster 4 profile, which may illustrate a successively maturing immune system. Longitudinal profiling was also performed in Liu et al., [25] studying the plasma proteome in 10 children between 9 months and 14 years, where the expression of 1747 proteins could be categorized into seven patterns of temporal changes, including a group of no change. Furthermore, they saw an age-effect for 970 of these proteins. [25] More recently, Lietzén et al. [26] found age to be the most dominant factor influencing plasma protein levels during the first 3 years of life in 14 children. Here, the 122 proteins with an age effect represented six temporal patterns.
Consistent with our observation, prominent changes in protein levels were seen during the first year of life. We additionally observed continued changes throughout the whole study period up to 5 years of age.
Insights into the longitudinal dynamics of proteins in early preterm life and up to 3 months of life were recently published. [4,27,28] Zhong et al. [28] profiled 448 proteins in serum samples from extremely preterm infants followed from birth up to term-equivalent age. Bennike et al., [27] studied the plasma proteome over the first week of life in full-term newborns, highlighting developmental changes in compo-nents of the complement system including C1QB, C4A, and C9 as also seen in our study, suggesting a continued role of the complement at later ages. Furthermore, in Olin et al., [4] 267 plasma proteins were profiled at birth and at 1, 4, and 12 weeks of age, with only few children sampled at 6 months of age. Although the sampled timeframe of these studies [4,27,28] is not comparable to ours, they provide observations of protein changes at earlier time points.
Lifestyle and environmental factors are known to influence both health and disease and various exposures may be reflected in changes in the plasma proteome. When we analyzed the effect of lifestyle, 59 of 97 proteins were altered over time within one or more of the lifestyle groups, and nearly half of these (28 out of 59) were altered independent of lifestyle group. There was an influence of lifestyle on a protein level shown by some proteins being changed uniquely within only one of the lifestyle groups ( Figure 3B). However, no single protein was identified that only showed changes in any of the three lifestyle groups.
This supports the notion that the changes are due to ontogeny and less so due to lifestyle differences ( Figure 3B). In addition, the SOTA clustering indicated more lifestyle-related differences ( Figure 3A). Interestingly, the highest number of proteins (n = 9) with a unique SOTA cluster assignment was found in the anthroposophic group ( Figure 3A).
However, further studies are required to determine the source of this variability and identify associations with exposures that are specifically related to the families' way of life and to clinical relevance.
A strength in our study is the unique, comparatively large, longitudinal approach in a birth cohort with well characterized different lifestyle groups [5] in which the same children were monitored four times, already from infancy up to 5 years of age. In addition, previously discussed studies, [2,3,25,26] have for method related reasons been relatively limited in sample size. In the present study, we applied an affinity proteomics approach [29] using a bead-based immunoassay with one of its major strengths being that it offers high sample throughput analysis, enabling in this case the large-scale profiling of 280 samples from 70 children. As the method only consumes a few microliters of sample, it is especially useful in settings such as those involving infants and young children, where limited sample volumes may be collected. This method has previously been applied in both pediatric and adult cohorts in other contexts such as childhood asthma and malaria, [12,30] liver and kidney disease, [11,31] nervous system/neurodegenerative disease, [32,33] and cancer. [34] Although our study included a large number of children, the number of children per lifestyle group was too small to allow for analysis stratified by exposure, such as vaccination and breast feeding or outcome such as sensitization to allergens. An association with sex in the development of plasma proteins cannot be ruled out; however, the proportions of females and males were reasonably balanced, also between the lifestyle groups ( Table 1). The method that was applied for plasma profiling used a single binder approach that generated relative levels of proteins, [13] and although this does allow for identification of longitudinal patterns, follow-up with methods for absolute quantification would be of interest to get further insights into the magnitudes of temporal changes and how these differ between proteins as well as for the establishment of reference intervals. Furthermore, multiple antibodies were used to target 73 of the 151 proteins that were measured. Based on a quality assessment of these antibodies, only supportive antibody pairs were included in the statistical analysis whereas those with low correlation were excluded. Possible explanations for differing results include that the epitopes of the multiple antibodies may differ and/or be affected by post translational modifications and thereby be differentially accessible for antibody recognition and binding as well as off-target binding. Among the 68 proteins that were significantly altered over time, 15 proteins were measured with more than one antibody, adding further technical support and confidence in the measurement of these proteins.
In conclusion, a majority (70%) of the 97 plasma proteins analyzed showed altered plasma levels during infancy up to 5 years of age. This

CONFLICT OF INTEREST
A. Scheynius is a member in the Joint Steering Committee for the Human Translational Microbiome Program at SciLifeLab/Karolinska Institutet together with Ferring Pharmaceuticals, Switzerland. The rest of the authors declare that they have no relevant conflicts of interest.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.