Impact of surface characteristics on the peri-implant microbiome in health and disease

Abstract

Background:

Since little is known about the impact of implant surface modifications on the peri-implant microbiome we aimed to examine peri-implant communities in various surface types in order to better understand the impact of these surfaces on the development of peri-implantitis.

Materials and methods:

106 systemically healthy individuals with anodized (AN), hydroxyapatite-coated (HA) or sandblasted acid-etched (SLA) implants that were >6 months in function were recruited and categorized into health (H) or peri-implantitis (PI). Peri-implant biofilm was analyzed using 16S rRNA gene sequencing and compared between health/disease and HA/SLA/AN using community-level and taxa-level metrics.

Results:

Healthy implants did not demonstrate significant differences in clustering, alpha- or beta-diversity based on surface modification. AN and HA surfaces displayed significant differences between health and peri-implantitis (p<0.05), however, such a clustering was not evident with SLA (p>0.05). AN and HA surfaces also differed in the magnitude and diversity of differences between health and PI. Six species belonging to the genera Shuttleworthia, Scardovia and Prevotella demonstrated lower abundances in AN implants with PI, and 18 species belonging to the genera Fretibacterium, Tannerella, Treponema and Fusobacterium were elevated, while in HA implants with PI, 20 species belonging to the genera Streptococcus, Lactobacillus, Veillonella, Rothia, and family Ruminococcaceae were depleted and Peptostreptococcaceae, Atopobiaceae, Veillonellaceae, Porphyromonadaceae, Desulfobulbaceae and order Synergistales were enriched.

Conclusion:

Within the limitations of this study, we demonstrate that implant surface can differentially modify the disease-associated microbiome, suggesting that surface topography must be considered in the multi-factorial etiology of peri-implant diseases.

One-sentence summary:

Implant surface topography is a modifier of the disease-associated peri-implant microbiome, and this impact varies widely between different surface modifications.

Introduction

Peri-implantitis is a biofilm-induced disease affecting the tissues that surround and support endosseous root-form implants¹. Nearly one in five implants are affected by peri-implantitis², with disease occurring as early as 2 years following functional loading³. Current treatment modalities, which are extrapolated from the natural dentition, do not predictably restore or reconstruct lost tissue⁴, and the risk of disease recurrence is 100%⁵. However, dental implants differ from teeth in important architectural aspects, namely, morphology, surface material, roughness and energy⁶. Therefore, elucidating the molecular etiology of peri-implant disease is critical to long-term success of therapeutic outcomes.

Most implants currently being used are engineered with moderately rough surfaces, with roughness levels ranging from 1.0 to 2.0 𝜇m⁷. However, with a view to improving osseointegration, additional surface modifications have been incorporated into implant design^7–9. These include machining, sand blasting, acid-etching, sintering, oxidizing, plasma-spraying, hydroxyapatite coating, laser-modification, or a combination of these procedures⁹.

Surface characteristics have been extensively studied with respect to their role in the incidence of peri-implant diseases, as well as their impact on the response to therapy^9–16. However, peri-implantitis is a dysbiosis-driven disease, and explicating their impact on the peri-implant microbiome is important to understanding disease etiology. Surface roughness and surface free energy have been identified as important determinants of biofilm composition in the peri-implant sulcus^{17, 18}. Additionally, several in vitro investigations reveal that P.gingivalis, Streptococcus pyogenes, S. mutans, A.naeslundii, F.nucleatum, and Lactobacillus salivarius adhere significantly more effectively to roughened titanium, either acid-etched or sand blasted surfaces^19–24. While some studies found no significant differences in microbial composition around diseased implants with different surfaces^{25, 26}, these studies were performed in dogs and limited to targeted pathogens that are not representative of natural pathobionts in that animal model

Sequence-based approaches have revolutionized our knowledge of various human habitats by elucidating factors that facilitate colonization and mobilization of these resident microbiota^{27, 28}. This approach has enabled us to establish that the peri-implant microbiome is distinct from the periodontal microbiome in critical ways^29–31, and to explicate the impact of environmental factors such as smoking on the peri-implant microbiome³². Therefore, the goal of this study was to examine the influence of implant surface topography on the peri-implant microbiome in both health and diseased conditions in humans using a comprehensive, open-ended, cultivation-independent approach to characterize and quantify the peri-implant microbiome.

Materials and Methods

Ethics statement

This project was approved by the Institutional Review Board (IRB) at the University of Michigan (Study ID: HUM00150546) and all the participants signed written informed consent before enrolling in the study.

Subject selection and recruitment:

Participants aged 18 years or older with peri-implantitis (PI) and healthy implants were recruited from the clinics at the University of Michigan School of Dentistry clinics. The participants were divided into three groups based on surface coatings of the implant i.e., anodized surface (AN), sandblasted and acid-etched surface (SLA), and hydroxyapatite coated surface (HA). Within each group, individuals were further divided into those with PI and those with healthy implants. Inclusion criteria for all participants were: systemic health (American Society of Anesthesiologists (ASA) I or II), and implant loading for at least 6 months. Exclusion criteria for all groups included heavy smokers (≥10 cigarettes/day), pregnant women, uncontrolled diabetes (≧7%), aggressive forms of periodontal disease, usage of antibiotic in 3 months prior to sample collection, radiation therapy in head and neck area within 2 years and current or previous use of steroids for cancer therapy. Implant health was characterized by an absence of visual signs of inflammation, bleeding on gentle probing and bone loss beyond physiological remodeling as outlined in the 2017 classification of periodontal and peri-implant diseases³³. Similarly, peri-implantitis was diagnosed based on visual signs of inflammation in the peri-implant soft tissue, combined with bleeding on probing and/or suppuration, bone loss beyond the radiographic bone level at the delivery of the implant-supported prosthetics reconstruction or evidence of bone loss ≥3 mm and/or probing depths ≥6 mm in conjunction with profuse bleeding in the absence of baseline data. Sites with history of intervention within the past 6 months were excluded from sample collection.

Sample size calculation:

In order to calculate sample size needed for each group to achieve statistically significant results, a power analysis was attempted by using the mean counts ± standard deviation (SD) of pathogenic microbes from previous studies. An electronic search using two databases (PubMed and Google Scholar) was conducted for all microbial studies on PI. However, most studies did not report the means or SD of the microbes³⁴, or reported them for as overall aggregates instead of group-wise data³⁵. Based on data from a single checkerboard DNA-DNA hybridization and real-time PCR data on Anodized TiUnite surface implants diagnosed with PI³⁶, a pilot study was planned to include 10 healthy implants and 30 implants diagnosed with PI for each implant surface design, giving a total of 120 implants for the entire study.

Sample collection, DNA isolation and sequencing:

The implant was isolated with cotton rolls, supra-mucosal biofilm or debris removed and dried with an air syringe; following which two sterile paper points were inserted into the sulcus until resistance was met and left in place for 30 seconds. The paper tips were placed into 2 mL sterile Eppendorf tubes filled with 1ml of DNA and RNA transport and storage medium^@. DNA extraction was performed using a magenetic bead-based kit^# on an automated platform. The V3-V4 region of the 16S rRNA gene was sequenced on the Illumina^® Miseq system using 250bp paired-end chemistry. The primer sequences have been previously published³⁷.

Bioinformatic analysis:

Amplicon Sequence Variants (ASVs) were inferred using the DADA2 v1.16 pipeline³⁸. Sequences were truncated based on quality plots, and filtered, dereplicated and denoised using standard parameters, following which chimeras were identified and removed. Paired ends of denoised sequences were then merged. In order to be retained in the dataset, the sequence had to be detected at least once in at least 5% of the samples. ASVs were assigned taxonomic identity using QIIME2 pre-fitted sklearn-based taxonomy classifier³⁹ by extracting the corresponding hypervariable region from the Human Oral Microbiome Database (HOMD v.15.22)⁴⁰. Alpha (within-group) diversity was computed using ACE⁴¹ and species diversity and richness using Shannon indices⁴². CSS (cumulative-sum-scaling) normalized OTU counts were used to evaluate beta diversity using Weighted and Unweighted Unifrac matrices with PhyloToAST v1.4⁴³. Linear Discriminant Analysis (LDA) was performed using the MASS package for R. The input for LDA was a matrix of variance-stabilized (arc-sin square root transformed) relative abundances of ASVs⁴⁴. MASS:lda provided singular value decomposition (svd) values, which were used to calculate the percent variance explained in each dimension. Wilk’s Lambda was used to test significance of LDA clustering. Linear discriminant analysis effect size (LEfSe)⁴⁵ was employed to identify drivers of community differences at the species level using a threshold of 2.0 for logarithmic linear discriminant analysis scores and an alpha value of 0.01 for the Kruskal-Wallis test. Differential abundance analysis of OTUs was carried out using DESeq2⁴⁶ and P-values adjusted for multiple testing (FDR<0.05, FDR-adjusted Wald Test). The ecological networks of the bacterial community that associated with health and disease were constructed using the SParse InversE Covariance Estimation for Ecological Association Inference (SPIEC-EASI)⁴⁷ association measure utilizing the Meinshausen and Bühlmann method with the help of NetComi v.1.1 package⁴⁸. Network clustering was performed using cluster_fast_greedy algorithm followed by cluster similarity assessment using Jaccard index and Adjusted Rand Index (ARI).

Results

In this study, 106 participants were recruited, including 30 healthy controls and 76 with peri-implantitis (PI). The healthy control was further divided into 3 subgroups of 10 AN, SLA and HA implants each, while the PI group comprised of 30 AN, 30 HA and 16 SLA implants. There were no demographic differences between healthy and PI groups, nor were there clinical differences between the AN, SLA and HA groups (Table 1). A total of 1,232,075 high quality sequences were analyzed and assigned to 487 OTUs with an average of 367 ± 114 OTUs in each sample.

Table 1:

Demographic and clinical characteristics of the sample population

	Health			Peri-implantitis
	Anodized	Hydroxyapatite	Sand blasted acid etched	Anodized	Hydroxyapatite	Sand blasted acid etched
Number of patients	10	10	10	30	30	16
Number of females	5	7	4	18	14	9
Age (mean ± SD)	61.7 ± 10.1	67.2 ± 10.2	66.7 ± 11.2	69.1 ± 8.1	67.1 ± 10.8	68.4 ± 10.5
BMI (mean ± SD)	28.4 ± 5	25.3 ± 6.6	26.6 ± 5.4	25.8 ± 5.4	26.2 ± 4.8	28.6 ± 5.9
Smoking status (never, former, current)	5,3,2	7,2,1	7,3,0	20,9,1	19,6,5	9,7,0
Diabetes (non-diabetic, controlled)	7,3	7,3	9,1	23,7	28,2	14,2
Periodontal status (Healthy, history of disease, active disease)	7,2,1	6,3,1	6,2,2	13,10,7	15,10,5	9,5,2
Implant location
Mandible (n, %)	3 (30%)	4 (40%)	7 (70%)	19 (63.3%)	14 (46.7%)	8 (50%)
Maxilla (n, %)	7 (70%)	6 (60%)	3 (30%)	11 (36.7%)	16 (53.3%)	8 (50%)
Anterior (n, %)	2 (20%)	1 (10%)	3 (30%)	1 (3.3%)	4 (13.3%)	1 (6.25%)
Posterior (n, %)	8 (80%)	9 (90%)	7 (70%)	29, 96.7	26 (86.7%)	15 (93.75%)
Prosthetic attachment (cemented)	7	7	6	23	18	12
Peri-implant soft tissue phenotype (thick)	7	8	7	18	18	12
Mean function time ± SD (months)	144.0 ±23.8	75.5 ± 12.9	121.6 ± 28.0	112.5 ± 29.7	67.1 ± 29.8	74.3 ± 29.7
Mean Probe Depth ± SD (mm)	4.3 ± 1.2	5.1 ± 2.1	4.7 ± 1.6	7.1 ± 1.71	7.53 ± 2.33	6.87 ± 1.20
Facial Bleeding on Probing
Mesial (n, %)	4 (40%)	6 (60%)	3 (30%)	15 (50%) / 7 (23.3%)	18 (60%) / 6 (20%)	12 (75%) / 1 (6.25%)
Mid (n, %)	0	1 (10%)	2 (20%)	15 (50%) / 10 (33.3%)	10 (33.3%) / 10 (33.3%)	8 (50%) / 3 (18.7%)
Distal (n, %)	3 (30%)	6 (60%)	1 (10%)	18 (60%) / 7 (23.3%)	16 (53.3%) / 7 (23%)	11 (68.7%) / 1 (6.25%)
Lingual Bleeding on Probing
Mesial (n, %)	4 (40%)	3 (30%)	2 (20%)	18 (60%) / 7 (23%)	17 (56.6%) / 6 (20%)	12 (75%) / 0
Mid (n, %)	0	2 (20%)	0	16 (53.3%) / 7 (23%)	11 (36.6%) / 5 (16.6%)	10 (62.5%) / 1 (6.25%)
Distal (n, %)	3 (30%)	3 (30%)	1 (10%)	20 (66.6%) / 7 (23%)	22 (73.3%) / 4 (13.3%)	12 (75%) / 1 (6.25%)

Implant surface characteristics do not impact the health-compatible peri-implant microbiome:

We first compared bacterial diversity between AN, SLA and HA healthy implants. LDA of weighted and unweighted Unifrac distances did not show significant clustering based on the surface modification (p=0.688, p=0.437 respectively) (Fig.1). Alpha diversity indices (ACE and Shannon) did not differ significantly between groups. Furthermore, LEfSe and DESeq did not identify species that were different between groups. Thus, in individuals with healthy implants, the peri-implant microbial community is not influenced by implant surface modifications.

Figure 1:

Linear Discriminant Analysis (LDA) of weighted and unweighted Unifrac distances clustered by implant surface modification are shown in Panels A and B respectively. The clustering was not statistically significant (p=0.688, p=0.437 respectively, Wilks’ lambda). Density curves of alpha diversity (ACE and Shannon) are shown in panels C and D (The peak indicates the median values for each group, and the x axis shows the data range). Again, the diversities were not significantly different.

Differences between healthy and PI-associated microbiomes vary by implant surface characteristics:

We then investigated whether implant surface characteristic or disease status was the bigger driver of microbial composition in established peri-implant disease. To do this, we first compared the biomes between health and PI agnostic of disease status, and found significant class separation (p=0.006, p=0.001, LDA of weighted and unweighted Unifrac distances respectively, Figures 2A and 2B). We then compared the AN, SLA and HA surfaces within the PI group. The clustering based on surface characteristic was non-significant (p=0.544, p=0.979, LDA of weighted and unweighted Unifrac distances respectively, Figures 2C and 2D). LefSe and DESeq failed to identify any taxa that could discriminate between disease-associated implant surfaces. This indicates that the effect of peri-implant disease on the peri-implant microbiome community is more pronounced than the effect of implant surface type. Although implant surface characteristic did not appear to be a determinant of the peri-implant microbiome in health or disease, each implant type demonstrated varying degrees and patterns of differences between health and disease (Figures 3A–3C). For instance, while anodized implants (p=0.023, p=0.011, LDA of weighted and unweighted Unifrac distances, respectively) and hydroxyapatite coated surfaces (p=0.049, p=0.007) displayed significant clustering between health and peri-implantitis, SLA implants did not demonstrate a significant difference between health and disease (p=0.916, p=0.829).

Figure 2:

Linear Discriminant Analysis (LDA) of weighted and unweighted Unifrac distances clustered by disease status are shown in Panels A and B and by implant surface modification in panels C and D. The microbiome demonstrated significant differences based on disease status (p=0.006, p =0.001 Wilks’ lambda), but not based on implant surface characteristics (p>0.05).

Figure 3:

Linear Discriminant Analysis (LDA) of weighted and unweighted Unifrac distances clustered by disease status for anodized implants are shown in Panels A and B, and organisms that were drivers of this clustering are shown in panel C. Similar metrics for hydroxyapatite implants are shown in panels D-F and for SLA implants in panels G-1.

However, even though AN and HA surfaces demonstrated significant differences between health and PI, they differed in the magnitude of these differences, as well as in the species that drove these differences. Despite demonstrating significant differences between health and PI, AN and HA surfaces varied in the extent of these differences and in the specific species responsible for these differences (Figure 4). For example, 14 differentially abundant species (padj < 0.05, log2FC < 20.0) were identified in AN implants with PI when compared to health, which included Porphyromonas gingivalis, Capnocytophaga sputigena, Prevotella spp., Treponema sp., Eubacterium spp., Johnsonella ignava, Leptotrichia hongkongensis and Corynebacterium durum. On the other hand, PI in HA implants was associated with higher levels of 24 species (padj < 0.05, log2FC < −7.0) belonging to Prevotella spp., Eubacterium spp., Desulfolobus sp, Streptococcus constellatus, Porphyromons sp., Cardiobacterium valvarum, Peptidophaga gingivicola, Lautropia mirabilis, Bulledia extruxta, Shuttleworthia satelles, Anaeroglobulus geminatus and Fretibacterium fastidiosum, and a lower abundance of Veillonella atypica when compared with health. In comparison, fewer significant differences (padj < 0.05) were observed with SLA implants. SLA implants in disease were associated with lower abundance of Prevotella histicola and Peptoniphilaceae bacterium HMT790 (log2FC < −23.0), but elevated levels of Anaeroloneae bacterium HMT439 and Capnocytophaga sp. (log2FC < 20.0).

Figure 4:

Operational taxonomic units (OTUs) that differed significantly between health and peri-implantitis of (a) Anodized, (b) HA and (c) SLA implants.

Additionally, ecological bacterial networks demonstrated different neighborhood hubs between health and PI for each implant surface (Figure 5A–C). Corroborating the LDA, Jaccard index and ARI demonstrated significant differences between health- and PI-associated central nodes and hubs irrespective of implant surface (p < 0.05). Higher percentage of negative associations as well as greater number of stronger positive associations were observed in health associated networks as compared to disease in all implant types. This suggests strong and stringent monitoring of the oral ecology in health. However, in line with the LefSe and DeSeq2 results, different network anchors were identified between health and PI in each implant surface type. For instance, Parvimonas micra (ASV461) and Prevotella pleuritidis (ASV265) anchored the PI-associated hub in AN implants, whereas Streptococcus (ASV56), Campylobacter gracilis (ASV75), Parvimonas (ASV125) and Prevotella (ASV135) were identified as anchors in healthy AN implants. Similarly, Streptococcus (ASV166), Actinomyces israelii (ASV70) and Fusobacterium nucleatum subsp. vincentii (ASV284) were influential in HA implants with PI, whereas the health-associated HA implant microbiome was anchored by Eubacterium brachy (ASV9), Schaalia sp. and Veillonella dispar (ASV276) as hub taxa. Similarly in SLA implants, Prevotella nigrescens (ASV8), Eubacterium brachy (ASV9), Streptococcus (ASV301) and Pseudoramibacter alactolyticus (ASV226) were identified as disease associated hub anchor, whereas Bacteroidales bacterium HMT274(ASV105) was a driver of health.

Figure 5:

Ecological networks between health and diseased conditions in (a) Anodized, (b) HA and (c) SLA implants. Each ASV is defined by a node, which is connected to other nodes with edges. Green and red edges represent positive and negative association between nodes respectively. The density of edges determines the strength of the association between nodes. Size of each node represents the node’s eigenvector centrality, and the color depicts the cluster.

Discussion

Dental implants restore impaired dental function, phonetic ability, and aesthetics in individuals with tooth loss. Since osseointegration is key to implant success, the implant surface has undergone significant modifications with the aim of accelerating and improving bone-to-implant contact, increasing cell viability and biocompatibility. This is most commonly achieved by altering its roughness through acid treatments, sandblasting, different mechanisms of oxidization, or a combination of these methods. Since microbial colonization can also be impacted by surface topography, we combined a cross-sectional clinical study design with an open-ended, comprehensive approach for bacterial identification and quantification to investigate the impact of this property on the peri-implant microbiome. To the best of our knowledge, this provides the first evidence of its kind from human studies.

In the present investigation, surface roughness was not a factor in determining the composition of the microbiome surrounding healthy implants. This was initially surprising, since it is known that bacterial colonization is highly influenced by topographical characteristics of the environment⁴⁹, and evidence from in vitro, animal and human studies demonstrate differences in bacterial adhesion and colonization rates among titanium discs with different topographies^50–52. However, the healthy implant is placed at or below the level of the alveolar crest and is sealed from the peri-implant sulcular environment by a soft tissue attachment apparatus. Therefore, it is logical that modifications to the surface of the implant body will not impact microbial colonization in the sulcus. The location of the implant-abutment connection, the material and the surface characteristics of the coronal structure are likely variables and further studies are warranted.

In line with this, implant surface topography was a factor in discriminating between health-compatible and disease-associated microbial profiles. Although AN and HA implants were microbially similar in health, following the establishment of PI, dysbiosis was more pronounced in HA implants than AN, with loss of several health-compatible species and enrichment of over 40 others. Although there are limited commonalities among the enriched species in peri-implantitis (PI) across different implant types, it is evident that the majority of enriched species in PI-associated implants belong to well-known oral pathogenic genera. For instance, in the case of AN implants, the genus Porphyromonas, Treponema, and Prevotella were prominently enriched. Porphyromonas spp. is widely recognized as a key causative bacterium in periodontal diseases and has also been found to be enriched in peri-implantitis compared to peri-mucositis or healthy conditions. HA implants, on the other hand, exhibited a broader range of PI-associated species compared to AN and SLA implants. These included members of the genus Veillonella and Eubacterium, which are also associated with periodontitis as accessory pathogens, in addition to those identified in AN implants.

HA-sprayed titanium implants are gaining popularity due to enhanced bone strength, early osseointegration and formation of a robust soft-tissue seal. However, early studies suggested that this surface modification could be susceptible to microbial colonization^{53, 54}, and that contamination can alter its surface characteristics, promoting bacterial adherence⁵⁵. Our study further bolsters this evidence and suggests that case selection is important for determining the type of implant for each patient.

An interesting finding was the lack of difference between health and peri-implantitis in SLA-modified implants. While it is possible to attribute this to the small sample size in this group, we have previously identified robust differences between peri-implant health and disease with a similar sample size³⁰. Several studies have raised the possibility that peri-implantitis has a heterogenous etiology; of which loss of foreign body equilibrium^{52, 56} and release of titanium particles and corrosion by-products within the surrounding tissue^{57, 58} have gained attention recently. It was not within the scope of this study to investigate these variables; however, our data suggest that factors other than dysbiosis might play a role in disease initiation.

In summary, within the limitations of a small sample size and a cross-sectional study design, our data suggest that surface topography is a modifier of the disease-associated peri-implant microbiome, and that the extent of this impact varies widely between the different modifications. The study underscores the multi-factorial etiology of peri-implant diseases and posits that implant surface topography might directly or indirectly influence susceptibility to disease.