Association of Surgical Hospitalization with Brain Amyloid Deposition: The Atherosclerosis Risk in Communities–Positron Emission Tomography (ARIC–PET) Study

Cognitive impairment after surgery and anesthesia in older adults is increasingly recognized as common and important. Accordingly, the American Society of Anesthesiologists (Schaumburg, Illinois) recently established a perioperative Brain Heath Initiative to focus awareness on this issue. The highest risk period for postoperative cognitive change appears to be days to weeks after surgery.^1–3  Long-term cognitive impairment or increased risk of dementia has also been described,^3–5  although the evidence regarding the putative role for surgery and anesthesia is conflicting.^6,7  For example, a recent study found differing associations between surgical hospitalization and dementia risk across clinical and registry-based cohorts.⁸  Risk factors for postoperative cognitive decline, such as critical illness,⁹  and delirium^10,11  have been identified, but the neurobiologic mechanisms that might underlie cognitive decline after surgical hospitalization are not clear. One hypothesis is that perioperative events promote the deposition of β -amyloid (amyloid), a protein that is thought to play a key role in the pathogenesis of Alzheimer’s disease.¹²  Preliminary support for this hypothesis has emerged from cell culture and animal studies, which suggest that surgery- and anesthesia-related mechanisms, including volatile anesthetics and inflammation, may cause an increased production and accumulation of brain amyloid.^13–15  However, data from humans are scarce, and to the best of our knowledge, there are no large studies examining cortical amyloid deposition in humans after surgery. Thus, there is a clear need to understand the molecular brain changes associated with undergoing surgery in middle and late adulthood, in particular the role of amyloid deposition. If surgery or perioperative events influence cortical amyloid deposition, this would have meaningful implications for clinical decision making and management, especially for older adults at risk for dementia.

We used data from a cohort of older adults enrolled in the Atherosclerosis Risk in Communities–Positron Emission Tomography (ARIC–PET) study¹⁶  to examine whether individuals hospitalized for a surgical procedure (henceforth referred to as “surgical hospitalization”) in the decades leading up to older adulthood had elevated cortical amyloid levels, as measured using florbetapir PET imaging. We hypothesized that surgical hospitalization, particularly for a procedure with moderate to high cardiac risk, is associated with increased cortical amyloid in late life.

ARIC is an ongoing community-based cohort study, which initially recruited 15,792 participants between 1987 and 1989 (visit 1) from four U.S. communities: Washington County, Maryland; Forsyth County, North Carolina; the northwestern suburbs of Minneapolis, Minnesota; and Jackson, Mississippi (African Americans only).¹⁷  As is illustrated in figure 1, participants were brought back for four additional visits until visit 5 (which occurred 2011 to 2013). At visit 5, approximately 2,000 participants (a group of participants with cognitive impairment and an age-matched group of cognitively normal participants) were selected to receive a brain magnetic resonance imaging (criteria for selection is outlined in the Supplemental Digital Content 1, https://links.lww.com/ALN/C304).¹⁸  Among these, 346 participants at three ARIC sites (Forsyth County, North Carolina; Jackson, Mississippi; and Washington County, Maryland) without dementia, heavy current alcohol use, renal dysfunction (with creatinine levels greater than 2 mg/dl), or prolonged QT-c interval (more than 450 ms) were selected to take part in ARIC–PET.¹⁶  Participant exclusion criteria are listed in figure 1, and a detailed study flowchart is provided in Supplemental Digital Content 2 (https://links.lww.com/ALN/C305). We excluded a small number of participants who did not meet ARIC–PET inclusion criteria (n = 3) or were missing important covariates (n = 4). We excluded another 26 participants who had missing International Classification of Disease, Ninth Revision codes because we could not ascertain the type of hospitalization. For the remainder of participants, we used a complete case analysis (with the underlying assumption of missing completely at random). All ARIC study protocols were approved by the Institutional Review Boards at each participating center: University of North Carolina at Chapel Hill, Chapel Hill, North Carolina; Johns Hopkins University, Baltimore, Maryland; University of Minnesota, Minneapolis, Minnesota; and University of Mississippi Medical Center, Jackson, Mississippi.

All hospital events (surgical and nonsurgical) were identified using active surveillance of local hospitals and annual telephone contact between the time-of-study baseline (visit 1) until the time of the PET scan. All hospital discharge codes were reviewed and abstracted by trained staff. Based on methods that have been described previously,¹⁹  we used International Classification of Disease, Ninth Revision, Clinical Modification codes and clinical classification software to categorize all surgical procedures documented during each hospitalization. International Classification of Disease, Ninth Revision, Clinical Modification codes have been validated as a method for the identification of hospital-based surgical procedures with generally good sensitivity and excellent specificity.²⁰  Surgical and other procedures were classified as either “low risk” or “moderate-to-high risk” by an experienced anesthesiologist according to 2014 American College of Cardiology/American Heart Association guidelines.²¹  Of note, the primary analyses included endoscopic procedures in the low-risk group based on the American College of Cardiology/American Heart Association guidelines. A complete list of surgical procedures listed according to their level of risk is provided in Supplemental Digital Content 3 (https://links.lww.com/ALN/C306). Participants with missing International Classification of Disease, Ninth Revision, Clinical Modification codes for one or more hospitalization (n = 26) were excluded unless the participant had a separate hospitalization that included a moderate/high surgery.

The outcome of interest was cortical amyloid deposition, as defined by florbetapir PET imaging. Compared with the gold standard (neuropathologic studies of postmortem amyloid burden), florbetapir PET has demonstrated excellent validity and reliability in distinguishing individuals with absent or sparse cortical amyloid from those with moderate to frequent plaques.²²  Visit 5 magnetic resonance imaging scans were analyzed at the ARIC Magnetic Resonance Imaging Reading Center (Mayo Clinic) using previously described methods.¹⁸  Florbetapir (amyloid) PET scans were performed within 1 yr of brain magnetic resonance imaging.¹⁶  PET imaging procedures are detailed in the Supplemental Digital Content 4 (https://links.lww.com/ALN/C307). We calculated a global measure of florbetapir uptake using a volume-dependent weighted average of the following regions: orbitofrontal, prefrontal, and superior frontal cortices; the lateral temporal, parietal, and occipital lobes; and the precuneus, the anterior cingulate, and the posterior cingulate. We used an automated area of cerebellar gray matter as a reference region.²³  In accordance with previously published methods,^16,24,25  elevated cortical amyloid, defined a priori as a global standardized uptake value ratio above the sample median (1.2), was prespecified as the primary outcome. Sensitivity analyses evaluated alternative thresholds for elevated amyloid (standardized uptake value ratios of more than 1.11²⁶  and more than 1.25) and examined global and brain region-specific amyloid standardized uptake value ratio as a continuous variable (log transformed to correct for skewness).

Participant age, sex, education, center (Maryland/North Carolina/Mississippi), and race (white/African American) were obtained at baseline from participant self-report. APOE allele status was determined using the TaqMan assay (Applied Biosystems, USA). We incorporated physiologic variables assessed during visit 5: body mass index, calculated from recorded height and weight (kg/m²); total cholesterol, measured using the enzymatic method²⁷ ; and high density lipoprotein cholesterol, calculated using Friedewald’s formula. Cigarette and alcohol use status (current/former/never) were determined based on participant self-report at visit 5.

We determined the presence/absence of the following medical conditions assessed at visit 5: hypertension, defined as antihypertensive medication use, or systolic or diastolic blood pressure of more than 140 mmHg and more than 90 mmHg, respectively; diabetes, defined as either participant report of a diabetes diagnosis from a physician, a fasting glucose level of at least 126 mg/dl, a nonfasting glucose level of at least 200 mg/dl, or current use of diabetes medication; coronary heart disease, adjudicated after visit 1 based on self-report or medical record evidence of previous myocardial infarction, coronary artery bypass graft, or angioplasty, or the presence of a myocardial infarction as determined by electrocardiogram; heart failure, defined as a previous heart failure–related hospitalization or heart failure medication use within the 2 weeks preceding the study visit; chronic obstructive pulmonary disease (COPD), defined based on participant report of previous COPD or emphysema diagnosis from a physician²⁸ ; and chronic kidney disease, defined based on estimated glomerular filtration rate, which was calculated using demographic characteristics and serum creatinine.²⁹ 

A statistical plan for these hypotheses was developed before accessing the data and approved by the ARIC study publications committee on February 14, 2017. We used multivariable logistic regression to examine the association of surgical hospitalization with cortical amyloid deposition. We compared participants with one or more previous surgical hospitalization with participants who were not previously hospitalized during the follow-up period (prespecified primary analysis). Additionally, we conducted a series of secondary analyses to determine how the use of alternative nonsurgery comparison groups may influence the findings. First, we compared participants with surgical hospitalizations with a group of hospitalized participants who did not undergo surgery (prespecified secondary analysis). Second, we compared participants with surgical hospitalization to the total group of participants without surgical hospitalization (i.e., the combined group of persons hospitalized without surgery and persons without previous hospitalization). Third, we examined the association of multiple surgical hospitalizations with cortical amyloid deposition by categorizing all participants into one of three groups based on the total number of surgical hospitalizations (zero, one, or at least two). Seventeen participants with one or more hospitalization with missing International Classification of Disease, Ninth Revision, Clinical Modification codes were excluded from the latter analysis. Findings from each of these secondary analyses are considered exploratory and should therefore be interpreted with caution.

To adjust for potential confounders, we used two regression models. The first model (model 1) adjusted for demographic factors (age, sex, education, race, center, and APOE ε4 status). We used a second model to account for the confounding effects of cardiovascular risk factors and chronic medical comorbidity.²⁴  This second model (model 2) adjusted additionally for late-life physiologic variables (body mass index, total cholesterol, and high-density lipoprotein), alcohol and cigarette use, and individual prevalent disease (hypertension, diabetes, coronary heart disease, heart failure, chronic kidney disease, and COPD) diagnosed at the time of visit 5. Post hoc analyses were conducted to examine whether findings differed when amyloid standardized uptake value ratio was parameterized as a continuous variable, and when the potential effects of selection into the ARIC–PET study were accounted for using inverse probability weighting (Supplemental Digital Content 5, https://links.lww.com/ALN/C308). Additionally, APOE ε4 status (defined as 0 vs. at least 1 ε4 alleles) was examined as an effect modifier. Additional sensitivity analyses examined the potential effect of reverse causation (i.e., amyloid brain changes increasing surgery risk) by excluding participants with an initial surgical hospitalization within the span of 5 yr before PET imaging and evaluated alternative thresholds for elevated amyloid. A minimum clinically meaningful odds ratio for elevated amyloid standardized uptake value ratio was not defined because of a lack of empirical data to guide this choice. The sample size for this analysis was derived based on available data from the ARIC–PET study.^16,24  A two-sided P value < 0.05 was used as the cutoff for statistical significance. Analyses were conducted using Stata, version 14 (StataCorp, USA).

Of the 313 participants included in the analytic sample, 56% (n = 175) were women, 40% (n = 125) were African American, 74% (n = 232) were cognitively normal, and 26% (n = 81) met criteria for mild cognitive impairment. Participants were 52.4 ± 5.2 yr of age (± SD) at study baseline and 76.0 ± 5.4 yr of age at the time of the PET scan. The average time between the baseline visit and PET scan was 25.2 ± 0.9 yr. Of the 226 (72%) participants who were hospitalized during the study period, 156 (69%) had one or more surgical hospitalizations, and 137 (61%) had one or more hospitalizations for a moderate to high-risk surgery. The average time between initial surgery and PET scan was 13.7 ± 7.4 yr; Supplemental Digital Content 6, https://links.lww.com/ALN/C309, histogram of the number of years between initial surgery and PET scan). As displayed in table 1, participants with a surgical hospitalization were slightly older and had greater prevalence of coronary heart disease compared with participants without a previous surgical hospitalization. Participants with one or more moderate-to-high-risk surgical hospitalization showed a similar pattern of a greater prevalence of heart disease and were less likely to possess a single APOE ε4 allele, compared with participants without surgical hospitalization. Group characteristics for participants with moderate to high-risk surgery are presented in Supplemental Digital Content 7 (https://links.lww.com/ALN/C310).

The prevalence of elevated amyloid levels among surgery, nonsurgery hospitalized, and nonhospitalized groups is displayed in figure 2A. In the primary analysis of 243 participants, which compared participants with one or more previous surgical hospitalization (87 of 156 [56%] amyloid-positive) with participants without previous hospitalization (45 of 87 [52%] amyloid-positive), there was no difference in odds of elevated brain amyloid after adjusting for demographic factors or after additionally adjusting for cardiovascular risk factors and prevalent disease (table 2). Similarly, in an analysis of 224 participants (excluding participants with only low-risk surgery), those with one or more moderate to high-risk surgeries (77 of 137 [56%] amyloid-positive) compared with participants without previous hospitalization (45 of 87 [52%] amyloid-positive) did not show increased odds of elevated brain amyloid after adjusting for demographic factors or after additionally adjusting for cardiovascular risk factors and prevalent disease (table 2). These results were similar in post hoc analyses that modeled amyloid as a continuous variable (Supplemental Digital Content 8, https://links.lww.com/ALN/C311), in analyses that used inverse probability weighting to account for selection (Supplemental Digital Content 9, https://links.lww.com/ALN/C312), in analyses that excluded individuals who underwent procedures that are generally performed with sedation alone (such as endoscopy and other minor procedures; Supplemental Digital Content 10, https://links.lww.com/ALN/C313), and in analyses that additionally adjusted for cognitive status and white matter hyperintensity volume (data not shown). There was no evidence for effect modification by APOE ε4 status (Supplemental Digital Content 11, https://links.lww.com/ALN/C314) or age (< 76 vs. ≥ 76, the sample median; Supplemental Digital Content 12, https://links.lww.com/ALN/C315).

In a prespecified secondary analysis (n = 226) that compared participants who had one or more surgical hospitalization (87 of 156 [56%] amyloid-positive) with participants who were hospitalized without surgery (27 of 70 [39%] amyloid-positive), surgical hospitalization was associated with a 2.10 greater odds (95% CI, 1.09 to 4.05) of elevated brain amyloid during late life after adjusting for demographic variables (small to medium effect size).³⁰  Similar results were observed after additionally adjusting cardiovascular risk factors (odd radio, 2.45; 95% CI, 1.13 to 5.33; table 3). Participants with one or more moderate-to-high-risk surgeries (77 of 137 [56%] amyloid-positive) also had a significantly higher odds of elevated brain amyloid during late life compared to participants who were previously hospitalized without moderate-to-high-risk surgery (37 of 89 [42%] amyloid-positive; table 3).

In an analysis of 313 participants that expanded the nonsurgery comparison group to include all participants (hospitalized and nonhospitalized) without a surgical hospitalization, participants with one or more surgical hospitalization (87 of 156 [56%] amyloid-positive) did not differ significantly from the nonsurgery group (72 of 157 [46%] amyloid-positive) with regard to odds of elevated amyloid (table 3). The results of post hoc analyses were largely similar when amyloid was modeled as a continuous variable (Supplemental Digital Content 13, https://links.lww.com/ALN/C316).

In an analysis of 280 participants not missing International Classification of Disease, Ninth Revision, Clinical Modification codes, participants with two or more surgical hospitalizations had 2.33 higher odds of elevated amyloid (95% CI, 1.04 to 5.24) during late life, compared with participants with no surgical hospitalizations (both hospitalized and nonhospitalized; table 4). Figure 2B displays the estimated probability of elevated amyloid according number of surgical hospitalizations.

The results were largely similar in analyses that excluded participants (n = 29) who had an initial surgery within 5 yr of PET imaging (Supplemental Digital Content 14, https://links.lww.com/ALN/C317) and in analyses that examined participants with their first surgical hospitalizations occurring proximal and distal to the time of PET imaging (distal defined as a surgical hospitalization at least 14 yr before PET imaging [median time]; Supplemental Digital Content 15, https://links.lww.com/ALN/C318). We also repeated analyses using alternative thresholds to define amyloid positive status.²⁶  Using a more liberal (standardized uptake value ratio of more than 1.11) and a more conservative (standardized uptake value ratio or more than 1.25) threshold, associations between surgical hospitalization and amyloid positive status were generally attenuated and were no longer statistically significant (Supplemental Digital Content 16 and 17, https://links.lww.com/ALN/C319 and https://links.lww.com/ALN/C320, respectively). Post hoc analyses that looked at the association of hospitalization with continuous markers of cortical amyloid across distinct brain regions are presented in Supplemental Digital Content 18 (https://links.lww.com/ALN/C321). The primary comparison showed no region-specific differences.

In a community sample of nondemented older adults enrolled in the ARIC–PET study, we did not find support for an association between past surgical hospitalization and elevated cortical amyloid levels in the primary analysis. Participants with one or more surgical hospitalizations did not differ significantly from participants without previous hospitalization with regard to odds of elevated amyloid, after adjusting for potentially confounding variables. The results were similar when the subset of participants with one or more moderate to high-risk surgeries was compared with participants without a previous hospitalization.

Few human studies have examined the association between surgery and brain amyloid deposition. A recent study measured florbetapir (amyloid) PET in a small number of patients 6 weeks and 1 yr after cardiac surgery. The authors found 1-yr increases in amyloid in the cardiac surgery group that were greater than that which has been reported previously in nonsurgery cohorts.³¹  However, this study was limited by a small sample size, the absence of a nonsurgery comparison group, and the exclusive focus on cardiac surgery patients limits the generalizability of these findings. Two human studies examined cerebral spinal fluid after surgery. Although these studies found a decreased amyloid/Tau ratio after surgery (indicative of greater Alzheimer’s disease burden), this change was largely due to an increase in Tau rather than changes in amyloid.^32,33 

Several studies have highlighted plausible mechanistic pathways through which the events associated with surgery may promote cortical amyloid deposition. First, animal models have indicated that common anesthetics, including sevoflurane and isoflorane, can increase the activation of caspases, leading to apoptosis and greater amyloid precursor protein processing.^13,34,35  Alzheimer’s disease transgenic mice have been shown to be particularly vulnerable to the effects of anesthesia because of an increased neuroinflammatory response.³⁴  Oligomerization of A β may also be increased after exposure to specific anesthetics such as isoflurane.¹⁵  Additionally, tissue injury resulting from surgery can generate damage-associated molecular patterns that can initiate a systemic inflammatory response. Systemic inflammation can subsequently trigger or exacerbate neuroinflammation, which itself is hypothesized to play a key role in amyloid deposition and cognitive change after surgery.³⁶  Additionally, sleep is commonly disrupted after surgery and is thought to play a key role in clearance of amyloid.^37,38  In spite of suggestive preclinical data, the results of the current study suggest that other mechanisms may be more important in the pathophysiology of cognitive change after surgery than brain amyloid deposition.

The results of the current study do not support the hypothesis that hospitalization for a surgical procedure during the mid- to late-life period increases the risk of elevated cortical amyloid. Surprisingly, we found the lowest rates of amyloid positivity among the group of participants who were previously hospitalized without surgery. It is unclear why participants with a previous nonsurgical hospitalization would have a lower prevalence of elevated amyloid than participants in the nonhospitalized group, because cardiovascular risk factors and other comorbidities have been associated with elevated brain amyloid.²⁴  One explanation for this unexpected finding may be selection bias. Given that persons with dementia were not included in the ARIC–PET study, it is possible that the comparatively higher prevalence of elevated amyloid among the nonhospitalized group represents a result of differential selection whereby persons without previous hospitalization (and without associated medical comorbidity) are more likely to remain nondemented in spite of a higher burden of amyloid and thus were not excluded from the ARIC–PET study.

Although the primary analyses did not support our hypothesis, we found some support for an association between previous surgical hospitalization and elevated cortical amyloid in secondary analyses that compared participants with surgical hospitalization to participants who were previously hospitalized without surgery and among participants who were hospitalized for two or more surgeries. However, the significant associations derived from secondary analyses were not robust to the use of alternative thresholds for defining elevated amyloid and therefore should be interpreted with caution. A recent study demonstrated that APOE genotype modified the association between hospitalization and dementia risk.⁸  However, our results do not support the hypothesis that APOE ε4 allele possession influences the relationship between surgical hospitalization and amyloid deposition.

The current study has several strengths, including the use of a unique, well characterized cohort followed over 24 yr, with measurement of hospitalization events, important cardiovascular risk factors, and medical comorbidity. There are also several limitations to consider. First, we found that the results varied based on the comparison group used. As discussed above, this may be due to a selection effect, whereby healthier nonhospitalized participants with elevated amyloid were able to remain nondemented and therefore be included in the original ARIC–PET study. Participants in the ARIC–PET cohort were selected to be dementia-free and able to undergo magnetic resonance and PET imaging. As such, this group may not be representative of the larger population of older adults, which is likely to have poorer health (see Supplemental Digital Content 19, https://links.lww.com/ALN/C322, participant characteristics stratified by inclusion in the ARIC–PET study). The selective attrition of persons with higher levels of medical comorbidity and the exclusion of persons with dementia from the ARIC–PET study, who presumably have a higher rate of surgical hospitalization and amyloid deposition, may have attenuated the associations found in the present study. Interpretation of the results may also be limited by potential bias caused by residual confounding from unmeasured covariates or subclinical disease.

Given that amyloid is known to increase for one to two decades and then plateau in older adults who go on to develop dementia,³⁹  the extended duration of time between surgery and amyloid assessment may mask surgery-related increases in amyloid deposition. As with other studies that measure brain amyloid levels, interpretation of the current results may also be limited by the lack of consensus regarding what characterizes a clinically meaningful threshold for (or difference in) brain amyloid levels. We have provided an estimate of the sample size needed to detect a group difference given the observed odds ratio of 1.32 from our primary analysis; however, it is unknown whether this observed effect is clinically meaningful (Supplemental Digital Content 20, https://links.lww.com/ALN/C323). The understanding of the effect of surgical hospitalization on amyloid deposition would be improved by the ability to assess how amyloid levels change after surgery. As such, the lack of baseline amyloid measurement represents a limitation of the current study. Another potential limitation, the use of International Classification of Disease, Ninth Revision, Clinical Modification codes to capture and categorize surgical procedures, may have introduced bias resulting from misclassification of the exposure variable. Last, use of these codes precluded the consideration of important exposures, such as type of anesthesia, postoperative complications, delirium, and drugs.

Despite these limitations, our results add to a growing body of evidence about the potential effects of surgery on the brain of aging adults. Although the primary analysis does not support an association between surgical hospitalization and elevated cortical amyloid, evidence from secondary analyses provides conflicting evidence. For older adults, preservation of cognitive and functional capacity are key patient-centered goals; however, there are a limited number of guidelines for evaluation and management strategies to preserve brain health after surgery. Thus, there is a need for more research to understand the neurobiologic mechanisms that underlie cognitive decline after surgery. Future studies with larger sample sizes, particularly studies that evaluate the level of cortical amyloid (or other potentially pathogenic proteins) both before and after surgery, may be especially important for providing more definitive evidence in this area.

The authors thank the staff and participants of the Atherosclerosis Risk in Communities study for their important contributions.

The Atherosclerosis Risk in Communities Study was carried out as a collaborative study supported by National Heart, Lung, and Blood Institute (Bethesda, Maryland) contract Nos. HHSN268201700001I, HHSN268201700002I, HHSN268201700003I, HHSN268201700005I, and HHSN268201700004I. Neurocognitive data was collected by grant Nos. U01 2U01HL096812, 2U01HL096814, 2U01HL096899, 2U01HL096902, and 2U01HL096917 from the National Institutes of Health (NHLBI, NINDS, NIA, and NIDCD; Bethesda, Maryland) and with previous brain magnetic resonance imaging examinations funded by grant No. R01-HL70825 from the National Heart, Lung, and Blood Institute. The ARIC–PET study is funded by National Institute on Aging (Bethesda, Maryland) grant No. R01AG040282. This study was also supported by contract Nos. K23AG064122 and T32 AG027668 (to Dr. Walker) and K24 AG052573 (to Dr. Gottesman) from the National Institute on Aging. Avid Radiopharmaceuticals (Philadelphia, Pennsylvania) provided the florbetapir isotope for the study but had no role in the study design or interpretation of results.

Dr. Gottesman serves as Associate Editor for Neurology. Dr. Knopman serves on a data safety monitoring board for the Dominantly Inherited Alzheimer Network study and is an investigator in clinical trials sponsored by Biogen (Cambridge, Massachusetts), Lilly Pharmaceuticals (Indianapolis, Indiana), and the Alzheimer’s Disease Cooperative Study. Dr. Wong receives funding from Avid/Lilly research collaboration, Roche Neuroscience (Basel, Switzerland), Lundbeck (Copenhagen, Denmark), Five Eleven Pharma (Philadelphia, Pennsylvania), and Cerveau (Knoxville, Tennessee) research collaboration, and his spouse is an intramural NIH scientist in a related field of the publication, but not a coauthor. Dr. Brown has consulted for and participates in a data-sharing agreement with Medtronic. The other authors declare no competing interests.