Abstract
Aims/Introduction
To prevent diabetic complications, strict glucose control and frequent monitoring of blood glucose levels with invasive methods are necessary. We considered the monitoring of tear glucose levels might be a possible method for non‐invasive glucose monitoring. To develop tear glucose monitoring for clinical application, we investigated the precise correlation between the blood and tear glucose concentrations.
Materials and Methods
A total of 10 participants and 20 participants with diabetes were admitted, and blood and tear samples were collected. Before statistical analysis, we eliminated tear samples contaminated with blood. We observed the daily blood and tear glucose dynamics, and carried out a random intercept model analysis to examine the association between the blood and tear glucose concentrations.
Results
Tear occult blood tests showed that the tear glucose concentrations and their variation increased in both participants with and without diabetes as contamination of blood increased. In both participants with and without diabetes, fluctuations of the plasma glucose concentrations were observed depending on the timing of collection of the samples, and the dynamics of the tear glucose concentrations paralleled those of the plasma glucose concentrations. The random intercept model analysis showed a significant association between the plasma and tear glucose concentrations in participants with diabetes (P < 0.001). This association still existed even after adjusting for the glycated hemoglobin levels and the prandial state (P < 0.001).
Conclusions
It is important to eliminate the tear samples contaminated with blood. Tear glucose monitoring might be a reliable and non‐invasive substitute method for monitoring the blood glucose concentrations for diabetes patients, irrespective of glycated hemoglobin levels and timing of sample collection.
Keywords: Glucose monitoring, Non‐invasive, Tear
It is important to eliminate the tear samples contaminated with blood. Tear glucose monitoring might be a reliable and non‐invasive substitute method for monitoring the blood glucose concentrations for diabetes patients, irrespective of glycated hemoglobin levels and the timing of sample collection.
Introduction
According to an estimate of the International Diabetes Federation, in 2017, there were 425 million diabetes patients aged 20–79 years, and 4 million deaths were caused by diabetes worldwide. It is predicted that by 2045, the number of diabetes patients would increase to 629 million 1 .
Clinical trials, such as the Diabetes Control and Complications Trial and United Kingdom Prospective Diabetes Study 33, have shown that strict glucose control is important to minimize the risk of microvascular complications, such as retinopathy, nephropathy and neuropathy 2 , 3 . The United Kingdom Prospective Diabetes Study 80 showed that intensive glucose control reduces the risk of macrovascular complications, such as myocardial infarction and diabetes‐related death, in the follow‐up period 4 .
To achieve strict glucose control, it is necessary for diabetes patients to measure their blood glucose levels frequently 5 . However, all currently available methods for blood glucose monitoring, including the finger prick method, are invasive. Therefore, non‐invasive blood glucose monitoring approaches have been investigated, but to date, none has been found to be a suitable substitute for direct glucose measurement in blood samples 6 . For example, one of the currently commercially available continuous glucose monitoring systems, which allows continuous measurement of the interstitial fluid glucose concentration, requires needling to place the sensor under the skin. In addition, for some of these systems, calibration to ensure the accuracy of the sensors necessitates measurement of the blood glucose levels (with invasive finger pricks) at least four times a day. Salivary glucose monitoring has been reported 7 ; however, the association between the blood glucose concentrations and salivary glucose concentrations in humans has not yet been sufficiently investigated, and further work is required to establish evidence of the association between blood glucose concentrations and salivary glucose concentrations.
Another possible non‐invasive glucose monitoring approach is tear glucose monitoring. Tear glucose concentrations in diabetes patients have been studied since 1937 8 , and many studies have reported that tear glucose concentrations are higher in diabetes patients than in participants with normal glucose tolerance 9 , 10 , 11 , 12 , 13 . In 1959, Lewis et al. 9 showed higher tear glucose concentrations in diabetes patients with hyperglycemia using enzymatic urine glucose test strips. In the 1980s, some studies used quantitative enzymatic methods to show the correlation between the blood glucose concentrations and tear glucose concentrations 11 , 12 . The dynamics of the tear glucose concentrations within a single day have been reported, with the trend shown to be the same as that of the blood glucose concentrations 12 , 13 .
One of the problems is that the order of the tear glucose concentrations varies widely among reported studies. For example, while the mean tear glucose concentration in participants with normal glucose tolerance was reported to be 0.032 mmol/L in one study 14 , another study reported a corresponding value of 0.160 mmol/L 11 . One possible reason is the difference in the type of stimulation (chemical, mechanical or non‐invasive) used among studies to induce tearing 15 . It is possible that the discrepancies in the tear glucose levels are derived from the differences in the methods of glucose measurement. In particular, the possibility of blood contamination is higher for tear samples with high glucose concentrations. It could also be attributable to the insufficient standardization of the conditions for tear sample collection, such as the time of day and the prandial state of the participants.
Some studies have shown linear regression formulas between blood glucose concentrations and tear glucose concentrations 13 , 16 . The correlations were statistically significant; however, the precision was insufficient to predict the blood glucose concentrations from tear glucose concentrations.
For clinical development of tear glucose monitoring as a tool for predicting the blood glucose concentrations, it is important to investigate the precise correlation between the blood glucose concentrations and tear glucose concentrations. In the present study, 10 participants without diabetes and 20 participants with diabetes were admitted to the University of Tokyo Hospital, Tokyo, Japan, overnight on two occasions, and blood and tear samples were collected (Figure 1). We observed the daily blood and tear glucose dynamics, and examined the association between the blood and tear glucose concentrations by random intercept model analysis with adjusting for confounders, using tear samples negative for occult blood.
Figure 1.
Schedule of sample collection. (a) Participants without diabetes. (b) Participants with diabetes. All participants were admitted to the hospital on two occasions according to this schedule.
Methods
This observational study was carried out with the approval of the research ethics committee of the University of Tokyo Hospital (11374‐(1)), and in accordance with the principles of the Declaration of Helsinki. Written informed consent to this study was obtained from all participants after explanation.
Participants
We recruited men without diabetes and men with type 2 diabetes mellitus aged ≥20 years for this study. The exclusion criteria, overall, included wearing of contact lenses, presence of ocular disorders, history of using eye drops, carriers of syphilis, hepatitis B virus, hepatitis C virus or human immunodeficiency virus and difficulty in the collection of tear samples. The exclusion criteria for the participants without diabetes included participants previously diagnosed as having diabetes and fasting blood glucose levels of ≥7 mmol/L. The exclusion criteria for the participants with diabetes included glycated hemoglobin (HbA1c) levels of <42 mmol/mol (6.0%) or ≥75 mmol/mol (9.0%), a history of receiving insulin and the presence of severe diabetic complications (Table S1). The reason for limiting our participants to those with HbA1c levels of <75 mmol/mol (9.0%) was that we were concerned about the safety of study participants with severe hyperglycemia during hospitalization. HbA1c (International Federation of Clinical Chemistry and Laboratory Medicine [IFCC]) was converted from HbA1c (National Glycohemoglobin Standardization Program) by IFCC = (10.93 × NGSP) − 23.50.
Schedule of sample collection
All samples were collected at the Phase 1 Unit, the University of Tokyo Hospital. The time course of sample collection for and participants with and without diabetes is shown in Figure 1. Both the participants with and without diabetes were admitted for 1 night to our hospital on two occasions within 1 month. On the day of admission, the participants without diabetes came to the hospital after fasting overnight, and the participants with diabetes took breakfast and their regular oral hypoglycemic medication(s) in the morning. Tear and blood samples were obtained before lunch, 2 h after lunch, before dinner, 2 h after dinner and before bedtime. On day 2 of admission, an oral glucose tolerance test was carried out in the participants without diabetes, and a meal tolerance test in the participants with diabetes, after overnight fasting. Tear and blood samples were collected before the glucose/meal load, and 30, 60 and 120 min after the load. Urine samples were also collected early in the morning before the glucose/meal load on day 2 of admission.
Meals during the admissions
The participants without diabetes consumed a normal diet, whereas the participants with diabetes consumed a diet restricted in calories to 7,531 kJ/day. All the participants were prohibited from having any food(s) other than that provided by the hospital. The meals in the second admission were the same as those in the first admission.
Blood and urine sampling and measurement
Blood samples were collected in vacuum‐sealed blood collection tubes. The glucose, HbA1c, insulin and creatinine concentrations in the blood samples were measured by SRL, Inc. (Tokyo, Japan). The urinary albumin and creatinine concentrations were also measured by SRL, Inc.
Tear sampling and measurement
Tear samples were collected with Microcaps micro capillaries (Drummond Scientific Co., Broomall, PA, USA). The tear fluid meniscus in the lower eyelid was gently touched with the micro capillary, and the tear fluid was drawn in by capillary action. Tear samples were collected from either the left or right eye. Chemical or mechanical tear stimulation was not used. We collected the tear samples immediately (within 10 min) after collecting the blood samples. All the tear samples were frozen at −80°C until laboratory measurement. A high‐pressure liquid chromatography method (e2695 Separations Module; Water Co., Tokyo, Japan) was used to measure the tear fluid glucose concentrations. The same standard glucose solutions were measured by both the methods used to measure the blood glucose concentrations (SRL, Inc.) and the tear glucose concentrations (high‐pressure liquid chromatography) to confirm the correlations of the results between the methods (Figure S1). If the volume of the tear samples was insufficient for measurement, samples were diluted to 20 μL with water before measurements. All the samples were divided into six portions, and the measurements were carried out on 6 days. On each day, calibration curves were drawn with the standard solutions before and after the measurements, to confirm the intraday and interday accuracy (Figure S2a–f).
Tear occult blood test
In many previous studies, it is known that the tear glucose concentrations vary widely 9 , 10 , 11 , 12 , 13 . Considering the possibility of blood contamination, we tested the samples for occult blood. To verify the validity of the application of urine test strips to tears, we used artificial tears (Artificial Tear Mytear Ophthalmic Solution; Senju Pharmaceutical Co., Ltd., Osaka, Japan) and hemoglobin standard substances (Interference Check A plus; Sysmex Corporation, Hyogo, Japan; Figure S3). The tear samples were first tested for occult blood using urine test strips (Terumo Uriace KC; Terumo Corp., Tokyo, Japan) and classified into five groups (occult blood −, +, ++, +++ and unmeasurable due to a shortage of samples; 56.3, 11.7, 20.0, 7.4 and 4.6%, respectively). The tear samples negative for occult blood were used for statistical analyses.
Statistical analysis
Age and laboratory data on admission, except overt proteinuria, were expressed as the mean ± standard deviation, and statistically compared between participants with and without diabetes by Welch’s t‐test, because the Kolmogorov–Smirnov test showed that some variables did not follow normal distribution (Table S2). We carried out Student’s t‐test for comparison between preprandial and 2‐h postprandial plasma and tear glucose levels on the first day between participants with and without diabetes. The daily plasma and tear glucose dynamics are shown, with the data expressed as the mean ± standard error of the mean. Receiver operating characteristic (ROC) analysis was carried out to verify its applicability to screening for hyperglycemia, which was defined as plasma glucose concentrations of >10 mmol/L in the present study. In this study, as 18 blood and tear samples were collected from each participant, the data showed a nested structure (i.e., the observations in each participant were not mutually independent). To properly analyze such nested data for evaluating the associations between the tear glucose and blood glucose concentrations, we used a random intercept model and random slope model (also called multilevel model, random effects model). The random intercept model decomposes the total variance of outcome into within‐participant fluctuations (i.e., residual variances) and between‐participant differences (i.e., random intercepts), whereas the random slope model decomposes the same variance into within‐participant fluctuations (i.e., residual variances) and between‐participant differences in terms of slopes (random slopes). Furthermore, confounders (HbA1c level and prandial state [preprandial/postprandial] of the participants) of the association between the plasma and tear glucose concentrations were included to the random intercept model. All the statistical analyses were carried out using the Statistical Package for Social Sciences (SPSS) version 22.0 (IBM Corp., Chicago, IL, USA). A significance level of α = 0.05 was set for all the analyses.
Results
Enrollment and outcomes
From 1 November 2016 to 31 January 2017, 10 participants without diabetes and 20 participants with diabetes were enrolled in the present study. The main exclusion criteria were unacceptably low or unacceptably high HbA1c levels (11 participants with diabetes), difficulty in the collection of tear samples (three participants without diabetes and three participants with diabetes), and withdrawal of consent by the participant (three participants with diabetes; Figure 2). All the participants completed the study without dropouts.
Figure 2.
Participant enrollment and outcomes. HbA1c, glycated hemoglobin.
Characteristics of the participants
The age and laboratory data of the participants are shown in Table 1. The mean age at study entry of the participants without diabetes was 31.2 years, whereas that of the participants with diabetes was 52.0 years (P < 0.001). The mean HbA1c level at the first admission of the participants without diabetes was 31.7 mmol/mol (5.1%), whereas that of the participants with diabetes was 54.1 mmol/mol (7.1%). The mean plasma insulin level of participants with diabetes was significantly higher than that of participants without diabetes (P < 0.05). Although it might be due to aging, the mean plasma albumin level of participants with diabetes was significantly lower than that of participants without diabetes (P < 0.05). The renal functions were not statistically different between participants with and without diabetes.
Table 1.
Age and laboratory data on admission
| Participants without diabetes | Participants with diabetes | P | |
|---|---|---|---|
| n | 10 | 20 | |
| Age (years) | 31.2 ± 0.4 | 52.0 ± 1.3 | <0.001 |
| HbA1c (mmol/mol) | 31.7 ± 3.4 | 54.1 ± 6.8 | <0.001 |
| HbA1c (%) | 5.1 ± 0.3 | 7.1 ± 0.6 | <0.001 |
| Plasma insulin (pmol/L) | 27.9 ± 15.7 | 107.6 ± 104.6 | 0.024 |
| Plasma albumin (g/dL) | 4.9 ± 0.3 | 4.5 ± 0.3 | 0.001 |
| Plasma creatinine (μmol/L) | 76.3 ± 7.3 | 82.9 ± 53.6 | 0.702 |
| Overt proteinuria, n (%) | 0 (0%) | 5 (25%) | 0.083 |
Each value, except overt proteinuria, shows the mean ± standard deviation. Plasma insulin levels were measured with blood samples before lunch on the day of admission. Overt proteinuria was determined by the urine albumin‐to‐creatinine ratio calculated by urine albumin and urine creatinine levels on the second day. Hba1c, glycated hemoglobin.
Tear occult blood test
In many previous studies, it is known that the tear glucose concentrations vary widely 9 , 10 , 11 , 12 , 13 . Considering the possibility of blood contamination, we tested the samples for occult blood. The tear samples were first tested for occult blood using urine test strips and classified into five groups (occult blood −, +, ++, +++ and unmeasurable due to shortage of samples; 56.3, 11.7, 20.0, 7.4 and 4.6%, respectively). As contamination of blood increased, the tear glucose concentrations and their variation increased in both participants with and without diabetes (Figures 3a,b; S6A‐D). Thus, we eliminated the tear samples positive for occult blood for the analyses, and used the tear samples negative for occult blood.
Figure 3.
(a,b) The box plots show the tear glucose concentrations in the five groups with occult blood testing. Open circles stand for outliers that are >1.5 and ≤3 box lengths above the 75% tile, and stars stand for outliers that are >3 box lengths above the 75% tile. (a) Participants without diabetes (occult blood −, +, ++, +++; n = 112, 17, 29, 7,15, respectively). (b) Participants with diabetes (occult blood −, +, ++, +++; n = 192, 46, 79, 33, 10, respectively). (c–f) Comparison of the plasma and tear glucose concentrations between participants without diabetes (open columns) and participants with diabetes (closed columns). Data are expressed as the mean ± standard error of the mean. (c) Preprandial plasma glucose concentrations at 12 and 18 o’clock on the day of admission (samples collected from participants without diabetes, n = 22; samples collected from participants with diabetes, n = 47). (d) Preprandial tear glucose concentrations at 12 and 18 o’clock on the day of admission (samples collected from participants without diabetes, n = 22; samples collected from participants with diabetes, n = 47). (e) The 2‐h postprandial plasma glucose concentrations at 14 and 20 o’clock on the day of admission (samples collected from participants without diabetes, n = 28; samples collected from participants with diabetes, n = 41). (f) The 2‐h postprandial plasma glucose concentrations at 14 and 20 o’clock on the day of admission (samples collected from participants without diabetes, n = 28; samples collected from participants with diabetes, n = 41). *P < 0.05, **P < 0.001. (g–j) Plasma (open circles) and tear (closed circles) glucose dynamics (samples collected from participants without diabetes, n = 7–18; samples collected from participants with diabetes, n = 19–24). Data are expressed as the mean ± standard error of the mean. Data are expressed as the mean ± standard error of the mean. (g) Participants without diabetes on the day of admission. (h) Participants with diabetes on the day of admission. (i) Participants without diabetes on day 2 of admission. (j) Participants with diabetes on day 2 of admission. AD, after dinner; AL, after lunch; BB, before bedtime; BD, before dinner; BL, before lunch.
Comparison of the plasma and tear glucose concentrations between participants with and without diabetes
The preprandial (before lunch and dinner on day 1 of both admissions) plasma glucose concentrations and tear glucose concentrations were significantly higher in the participants with diabetes than in the participants without diabetes (P < 0.001 and P < 0.05, respectively; Figure 3c,d). In addition, the 2‐h postprandial (2‐h after lunch and dinner on day 1 of both admissions) plasma glucose concentrations and tear glucose concentrations were also significantly higher in the participants with diabetes than in the participants without diabetes (P < 0.001 and P < 0.05, respectively; Figure 3e,f). The mean ± standard deviation of the overall value in the entire participant population and values in individuals of the tear glucose concentrations‐to‐plasma glucose concentrations ratio are shown in Figure S4.
Daily tear and blood glucose dynamics
The plasma and tear glucose dynamics are shown in Figure 3g–j. In both participants with and without diabetes, the plasma glucose concentrations fluctuated according to the timing of the sample collection, such as whether the samples were collected in the preprandial state, postprandial state or glucose‐loaded state of the participants. The dynamics of the tear glucose concentrations generally paralleled those of the plasma glucose concentrations.
ROC analysis
As the area under the curve of the ROC curve was 0.777, we determined that tear glucose measurement could detect plasma glucose concentrations of >10 mmol/L (Figure S5).
Random intercept model and random slope model analysis for examining the association between the plasma glucose concentrations and tear glucose concentrations
The random intercept model analysis showed that there was no association between the plasma glucose concentrations and tear glucose concentrations in participants without diabetes (Figure 4a), but there was a tendency of association between the plasma glucose concentrations and tear glucose concentrations in participants with diabetes (P = 0.054; Figure 4b; Table 2). There was also no association between the plasma glucose concentrations and tear glucose concentrations in participants without diabetes after logarithmic transformation of the tear glucose concentrations (Figure 2c), but there was a significant association between the plasma glucose concentrations and tear glucose concentrations in participants with diabetes (P < 0.001, Figure 4d; Table 2). In addition, there was a significant association between the plasma glucose concentrations and tear glucose concentrations, as well as between the plasma glucose concentrations and logarithmically transformed values of the tear glucose concentrations in participants with diabetes by random slope model analysis (Table 2). These results suggest that the association between the tear glucose concentrations and plasma glucose concentrations were stronger in the participants with diabetes whose plasma glucose concentrations were high. The random intercept model with the confounders (HbA1c level, prandial state [preprandial/postprandial] of the participants) to determine the association between the plasma glucose concentrations and logarithmically transformed values of the tear glucose concentrations showed that the association remained statistically significant even after adjustments for such confounders (P = 0.002), and the random effects of intercepts were significantly different among participants with diabetes (P = 0.015; Table 3).
Figure 4.
(a,b) Correlation charts and regression lines between the plasma and tear glucose concentrations. (a) Samples collected from participants without diabetes; n = 112. (b) Samples collected from participants with diabetes; n = 192. (c,d) The correlation charts and regression lines between plasma glucose concentrations and logarithmically transformed values of the tear glucose concentrations. (c) Samples collected from participants without diabetes; n = 112. (d) Samples collected from participants with diabetes; n = 192. All regression lines and P‐values were calculated by random intercept model analysis. NS, not significant.
Table 2.
Random intercept model and a random slope model for the tear glucose concentrations to examine its association with the plasma glucose concentrations in participants with diabetes
| Response variable | Tear glucose concentration | log10 (Tear glucose concentration) | ||||||
|---|---|---|---|---|---|---|---|---|
| Random intercept model | ||||||||
| Fixed effect | stimate | t | P | 95% CI | Estimate | t | P | 95% CI |
| Intercept | 0.096 | 1.289 | 0.201 | −0.052, 0.243 | −1.289 | −11.592 | <0.001 | −1.509, −1.068 |
| Plasma glucose (mmol/L) | 0.012 | 0.006 | 0.054 | −0.000, 0.024 | 0.041 | 4.468 | <0.001 | 0.023, 0.059 |
| Random effect | stimate | z | P | 95% CI | Estimate | z | P | 95% CI |
| Residual (level 1) | 0.036 | 9.183 | <0.001 | 0.029, 0.035 | 0.079 | 9.205 | <0.001 | 0.064, 0.098 |
| Intercept (level 2) | 0.026 | 2.403 | 0.016 | 0.011, 0.058 | 0.597 | 2.476 | 0.013 | 0.027, 0.132 |
| Response variable | Tear glucose concentration | log10 (Tear glucose concentration) | ||||||
|---|---|---|---|---|---|---|---|---|
| Random slope model | ||||||||
| ixed effect | Estimate | t | P | 95% CI | Estimate | t | P | 95% CI |
| Intercept | 0.047 | 0.728 | 0.467 | −0.080, 0.173 | −1.375 | −14.026 | <0.001 | −1.569, −1.182 |
| Plasma glucose (mmol/L) | 0.016 | 2.265 | 0.026 | 0.002, 0.030 | 0.049 | 4.431 | <0.001 | 0.027, 0.071 |
| Random effect | Estimate | z | P | 95% CI | Estimate | z | P | 5% CI |
| Residual (level 1) | 0.037 | 9.146 | <0.001 | 0.030, 0.046 | 0.084 | 9.127 | <0.001 | 0.068, 0.104 |
| Slope (level 2) | 0.000 | 2.182 | 0.029 | 0.000, 0.000 | 0.001 | 2.239 | 0.025 | 0.000, 0.001 |
Fixed effect: means of the intercept and regression coefficients. Random effect: variances of the intercept and regression coefficients. Fixed effect: means of the intercept and regression coefficients. Random effect: variances of the slope and regression coefficients. CI, confidence interval.
Table 3.
Random intercept model for logarithmically transformed values of the tear glucose concentrations to examine its association with the plasma glucose concentrations in participants with diabetes by including glycated hemoglobin and the prandial state (preprandial/postprandial) of the participants as confounders
| Fixed effect | Estimate | t | P | 95% CI |
|---|---|---|---|---|
| Intercept | −1.299 | −2.600 | 0.018 | −2.347, −0.251 |
| Plasma glucose (mmol/L) | 0.038 | 3.194 | 0.002 | 0.015, 0.062 |
| HbA1c (mmol/mol) | 0.000 | 0.031 | 0.976 | −0.019, 0.020 |
| Preprandial (0) or Postprandial (1) | 0.113 | 0.418 | 0.677 | −0.042, 0.064 |
| Random effect | Estimate | z | P | 95% CI |
|---|---|---|---|---|
| Residual (level 1) | 0.080 | 9.172 | <0.001 | 0.064, 0.099 |
| Intercept (level 2) | 0.065 | 2.429 | 0.015 | 0.029, 0.146 |
Fixed effect: means of the intercept and regression coefficients. Random effect: variances of the intercept and regression coefficients. CI, confidence interval; HbA1c, glycated hemoglobin.
Discussion
In the present study, we showed the existence of a significant association between the plasma glucose concentrations and tear glucose concentrations, with the association remaining significant even after adjustments for the HbA1c level and prandial state (preprandial/postprandial) of the participants by random intercept model analysis, which can evaluate both interpersonal and intrapersonal association. These results suggest that tear glucose measurements might be useful, irrespective of the characteristics of the population surveyed (e.g., diabetes control status) and the timing of sample collection (e.g., preprandial/postprandial).
The strength of the present study lies in the fact that it is the first to use random intercept model analysis to determine the association between the plasma glucose concentrations and tear glucose concentrations using tear samples not contaminated with blood. Furthermore, we carried out the measurements in one of the largest number of samples (n = 304) recorded to date, using a sensitive and reliable high‐pressure liquid chromatography method.
The limitations of the present study were that all of the participants were men, and the influence of sex on the association between the plasma and tear glucose concentrations was not investigated. Furthermore, the mean age of the participants with diabetes in the present study was significantly higher than that of the participants without, and it is possible that some of the relevant variables were influenced by age.
The association between the tear glucose concentrations and blood glucose concentrations were known to be insufficient to allow tear glucose monitoring to be used as a substitute for the invasive blood glucose monitoring. Various reasons for the insufficient association are speculated, including those related to the method of tear collection, method of stimulation of tearing, contamination of blood, method of measurement of the tear glucose concentrations and the effects of diabetes itself 17 . In this study, we used not only a non‐invasive tear collection method without stimulation of tears, but also the first tear occult blood testing to eliminate the tear samples positive for occult blood.
In addition to the methodological issues described above, the effects of diabetes itself should be considered. Hyperglycemia in diabetes patients causes microvascular damage and nerve damage 18 , 19 , and such damage to the vasculature supplying blood to the eyes and the nerves of the lacrimal reflex arc could alter tear production. For example, dry eyes can result because of reduced tear production. In fact, the prevalence of dry eyes is known to increase with HbA1c levels among participants with diabetes 20 . To rule out the effects of diabetes severity and prandial state, we used random intercept model analysis to eliminate the confounding effects of HbA1c and prandial state (preprandial/postprandial). The random effect of intercepts was significantly different among participants with diabetes. This result suggests that the associations between plasma and tear glucose concentrations vary by participants, and a tear glucose monitoring device requires calibrations to predict blood glucose levels properly.
The random intercept model and charts (Figure 4a–d) showed that the correlation between the tear glucose concentrations and plasma glucose concentrations were stronger in hyperglycemia states. In contrast, the correlation was not very good in normoglycemia states. In addition, the ROC analysis showed that tear glucose measurement could detect hyperglycemia without the need for calibrations. Therefore, when used for screening tests, tear glucose measurement can be expected, as a non‐invasive test that does not require calibrations. In contrast, for the purpose of daily blood glucose control in diabetes patients, there is a possibility of monitoring the blood glucose levels through non‐invasive tear glucose measurements without calibrations, if calibrations for correcting individual differences are carried out first. This is different from the case of continuous glucose monitoring systems, which require calibrations for correcting deterioration of the sensors throughout the measurements. In future studies, we propose to improve the accuracy of tear glucose measurements by devising better methods for collection and calibration.
Furthermore, we created models incorporating logarithmically transformed values of the tear glucose concentrations, which have scarcely been reported before. We transformed the data of tear glucose concentrations to account for normality assumption in applying the random intercept and slope model (Figure 4a–d; Table 2). The correlation charts and regression lines (Figure 4a–d) of the logarithm models appeared to be better.
The source of glucose in tears remains unclear. There are few reports about the existence of glucose transporters in the tear glands. The constitutive glucose transporter, glucose transporter 1, has been reported to be expressed in the apical corneal epithelium 21 . The sodium–glucose cotransporter, sodium–glucose cotransporter 1, has been reported to be expressed on the apical side of the bulbar and palpebral conjunctiva 22 , 23 . Although the roles of these transporters are still unclear, leakage of glucose from the interstitial space into the tear fluid cannot reach an equilibrium state, because tear fluid always flows out, contrary to the case for leakage of glucose from the blood vessels into the interstitial fluid.
Recently, it has been reported that tear fluid contains many kinds of metabolites, including glucose 24 , 25 . Some of these metabolites could serve as biomarkers of diseases (e.g., Alzheimer’s disease), paving the way for the development of a non‐invasive method for the diagnosis of these diseases 26 , 27 . Thus, tears appear to represent a very attractive material for diagnosis. In the future, we propose to measure other biomarkers associated with diabetes in tear samples.
In contrast, an ophthalmic lens that can measure the glucose concentrations in tears has been reported. This is a very attractive device, because it can measure the glucose concentrations continuously while eliminating the effects of the collecting method used 28 , 29 , 30 , 31 , 32 , 33 , and is expected to be an alternative device of continuous glucose monitoring systems, measuring interstitial fluid glucose concentrations. Although the sensitivity of these sensors is such that even glucose concentrations as low as those in tears can be measured, they are not yet suitable for use in clinical studies at this stage.
It has been known for a long time that there is an association between blood glucose and tear glucose concentrations 8 , 9 , 10 , 11 , 12 , 13 , 14 , 15 , 16 , 17 , but the present study revealed three new findings. The first is that there are individual differences in the association between blood glucose and tear glucose concentrations. The second is that this association still existed after adjusting for the HbA1c levels and the prandial state. Finally, contamination of the tear samples with blood led to a false elevation of the tear glucose concentration. The present results suggested the feasibility of measurement of biomarkers in tears, which could be collected non‐invasively. In the future, investigating biomarkers, including glucose in tears, is thought to be required toward the realization of non‐invasive measuring.
Disclosure
The authors declare no conflict of interest.
Supporting information
Figure S1 | Correlation charts between the glucose concentrations of standard solutions measured by both the methods used to measure the blood glucose concentrations (SRL, Inc.) and tear glucose concentrations (high‐pressure liquid chromatography).
Figure S2 | Calibration curves of high‐pressure liquid chromatography measurement before measurement (black line) and after measurement (red line). (a) Day 1, (b) day 2, (c) day 3, (d) day 4, (e) day 5 and (f) day 6.
Figure S3 | Discoloration of test strips depending on the hemoglobin concentration (0, 0.06, 0.15 and 0.75).
Figure S4 | The mean ± standard deviation of the total value in the entire participant population and values in individuals of the tear glucose concentrations‐to‐plasma glucose concentrations ratio.
Figure S5 | Receiver operating characteristic curve to detect plasma glucose concentrations of >10 mol/L. The area under the curve for this discrimination was 0.777.
Figure S6 | (a,b) Correlation charts between the plasma and tear glucose concentrations before excluding samples with occult blood (black, blue, yellow, red and green circles stand for –, +, ++, +++ and unmeasurable, respectively). (a) Samples collected from participants without diabetes (n = 180) and (b) samples collected from participants with diabetes (n = 360). (c,d) The correlation charts between plasma glucose concentrations and logarithmically transformed values of the tear glucose concentrations before excluding samples with occult blood (black, blue, yellow, red and green circles stand for –, +, ++, +++ and unmeasurable, respectively). (c) Samples collected from participants without diabetes (n = 180) and (d) samples collected from participants with diabetes (n = 360).
Table S1 | Exclusion criteria.
Table S2 | Kolmogorov–Smirnov test for validation of normal distribution.
Acknowledgments
We thank all the staff members of the Phase 1 Unit, the University of Tokyo Hospital, for their excellent assistance. This work was supported by a grant for Seed‐stage Technology‐based Startups from the New Energy and Industrial Technology Development Organization (NEDO).
J Diabetes Investig 2021; 12: 266–276
Contributor Information
Naoto Kubota, Email: nkubota-tky@umin.ac.jp.
Toshiya Sakata, Email: sakata@biofet.t.u-tokyo.ac.jp.
Takashi Kadowaki, Email: kadowaki-3im@h.u-tokyo.ac.jp.
References
- 1. International Diabetes Federation . IDF Diabetes Atlas 2017 . Available from: https://diabetesatlas.org/ Accessed April 1, 2019.
- 2. The Diabetes Control and Complications Trial (DCCT) Research Group. The absence of a glycemic threshold for the development of long‐term complications: the perspective of the Diabetes Control and Complications Trial. Diabetes 1996; 45: 1289–1298. [PubMed] [Google Scholar]
- 3. UK Prospective Diabetes Study (UKPDS) Group . Intensive blood‐glucose control with sulphonylureas or insulin compared with conventional treatment and risk of complications in patients with type 2 diabetes (UKPDS 33). Lancet 1999; 354: 837–853. [PubMed] [Google Scholar]
- 4. Holman RR, Paul SK, Bethel MA, et al 10‐year follow‐up of intensive glucose control in type 2 diabetes. N Engl J Med 2008; 359: 1577–1589. [DOI] [PubMed] [Google Scholar]
- 5. American Diabetes Association . Standards of medical care in diabetes–2014. Diabetes Care 2014; 37: S14–S80. [DOI] [PubMed] [Google Scholar]
- 6. Khalil OS. Non‐invasive glucose measurement technologies: an update from 1999 to the dawn of the new millennium. Diabetes Technol Ther 2004; 6: 660–697. [DOI] [PubMed] [Google Scholar]
- 7. Arakawa T, Kuroki Y, Nitta H, et al Mouthguard biosensor with telemetry system for monitoring of saliva glucose: a novel cavitas sensor. Biosens Bioelectron 2016; 84: 106–111. [DOI] [PubMed] [Google Scholar]
- 8. Michail D, Vancea P, Zolog N. Lacrimal elimination of glucose in diabetic patients. C R Soc Biol Paris 1937; 125: 194–195. [Google Scholar]
- 9. Lewis JG, Stephens PJ. Tear glucose in diabetics. Br J Ophthalmol 1958; 42: 754–758. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. van Haeringen NJ, Glasius E. Collection method dependant concentrations of some metabolites in human tear fluid, with special reference to glucose in hyperglycaemic conditions. Albrecht Von Graefes Arch Klin Exp Ophthalmol 1977; 202: 1–7. [DOI] [PubMed] [Google Scholar]
- 11. Sen DK, Sarin GS. Tear glucose levels in normal people and in diabetic patients. Br J Ophthalmol 1980; 64: 693–695. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Daum KM, Hill RM. Human tear glucose. Investig Ophthalmol Vis Sci 1982; 22: 509–514. [PubMed] [Google Scholar]
- 13. Lane JD, Krumholz DM, Sack RA, et al Tear glucose dynamics in diabetes mellitus. Curr Eye Res 2006; 31: 895–901. [DOI] [PubMed] [Google Scholar]
- 14. Taormina CR, Baca JT, Asher SA, et al Analysis of tear glucose concentration with electrospray ionization mass spectrometry. J Am Soc Mass Spectrom 2007; 18: 332–336. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Zhang J, Hodge W, Hutnick C, et al Noninvasive diagnostic devices for diabetes through measuring tear glucose. J Diabetes Sci Technol 2011; 5: 166–172. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Baca JT, Taormina CR, Feingold E, et al Mass spectral determination of fasting tear glucose concentrations in nondiabetic volunteers. Clin Chem 2007; 53: 1370–1372. [DOI] [PubMed] [Google Scholar]
- 17. Baca JT, Finegold DN, Asher SA. Tear glucose analysis for the noninvasive detection and monitoring of diabetes mellitus. Ocul Surf 2007; 5: 280–293. [DOI] [PubMed] [Google Scholar]
- 18. Writing Team for the Diabetes Control and Complications Trial/Epidemiology of Diabetes Interventions and Complications Research Group . Effect of intensive therapy on the microvascular complications of type 1 diabetes mellitus. JAMA 2002; 287: 2563–2569. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. King GL, Brownlee M. The cellular and molecular mechanisms of diabetic complications. Endocrinol Metab Clin North Am 1996; 25: 255–270. [DOI] [PubMed] [Google Scholar]
- 20. Kaiserman I, Kaiserman N, Nakar S, et al Dry eye in diabetic patients. Am J Ophthalmol 2005; 139: 498–503. [DOI] [PubMed] [Google Scholar]
- 21. Kumagai AK, Glasgow BJ, Pardridge WM. GLUT1 glucose transporter expression in the diabetic and nondiabetic human eye. Invest Ophthalmol Vis Sci 1994; 35: 2887–2894. [PubMed] [Google Scholar]
- 22. Turner HC, Alvarez LJ, Bildin VN, et al Immunolocalization of Na‐K‐ATPase, Na‐K‐Cl and Na‐glucose cotransporters in the conjunctival epithelium. Curr Eye Res 2000; 21: 843–850. [DOI] [PubMed] [Google Scholar]
- 23. Hosoya K, Lee VH, Kim KJ. Roles of the conjunctiva in ocular drug delivery: a review of conjunctival transport mechanisms and their regulation. Eur J Pharm Biopharm 2005; 60: 227–240. [DOI] [PubMed] [Google Scholar]
- 24. Galbis‐Estrada C, Martinez‐Castillo S, Morales JM, et al Differential effects of dry eye disorders on metabolomic profile by 1H nuclear magnetic resonance spectroscopy. Biomed Res 2014; 2014: 542549. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Zhou L, Zhao SZ, Koh SK, et al In‐depth analysis of the human tear proteome. J Proteomics 2012; 75: 3877–3885. [DOI] [PubMed] [Google Scholar]
- 26. Gergő Kalló G, Emri M, Varga Z, et al Changes in the chemical barrier composition of tears in Alzheimer’s disease reveal potential tear diagnostic biomarkers. PLoS One 2016; 11: e0158000. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Wood H. Alzheimer disease: could tear proteins be biomarkers for Alzheimer disease? Nat Rev Neurol 2016; 12: 432. [DOI] [PubMed] [Google Scholar]
- 28. Alexeev VL, Das S, Finegold DN, et al Photonic crystal glucose‐sensing material for noninvasive monitoring of glucose in tear fluid. Clin Chem 2004; 50: 2353–2360. [DOI] [PubMed] [Google Scholar]
- 29. Badugu R, Lakowicz JR, Geddes CD. Ophthalmic glucose monitoring using disposable contact lenses–a review. J Fluoresc 2004; 14: 617–633. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. March WF, Mueller A, Herbrechtsmeier P. Clinical trial of a noninvasive contact lens glucose sensor. Diabetes Technol Ther 2004; 6: 782–789. [DOI] [PubMed] [Google Scholar]
- 31. Domschke A, March WF, Kabilan S, et al Initial clinical testing of a holographic non‐invasive contact lens glucose sensor. Diabetes Technol Ther 2006; 8: 89–93. [DOI] [PubMed] [Google Scholar]
- 32. Yao H, Shum AJ, Cowan M, et al A contact lens with embedded sensor for monitoring tear glucose level. Biosens Bioelectron 2011; 26: 3290–3296. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33. Chu MX, Shirai T, Takahashi D, et al Biomedical soft contact‐lens sensor for in situ ocular biomonitoring of tear contents. Biomed Microdevices 2011; 13: 603–611. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Figure S1 | Correlation charts between the glucose concentrations of standard solutions measured by both the methods used to measure the blood glucose concentrations (SRL, Inc.) and tear glucose concentrations (high‐pressure liquid chromatography).
Figure S2 | Calibration curves of high‐pressure liquid chromatography measurement before measurement (black line) and after measurement (red line). (a) Day 1, (b) day 2, (c) day 3, (d) day 4, (e) day 5 and (f) day 6.
Figure S3 | Discoloration of test strips depending on the hemoglobin concentration (0, 0.06, 0.15 and 0.75).
Figure S4 | The mean ± standard deviation of the total value in the entire participant population and values in individuals of the tear glucose concentrations‐to‐plasma glucose concentrations ratio.
Figure S5 | Receiver operating characteristic curve to detect plasma glucose concentrations of >10 mol/L. The area under the curve for this discrimination was 0.777.
Figure S6 | (a,b) Correlation charts between the plasma and tear glucose concentrations before excluding samples with occult blood (black, blue, yellow, red and green circles stand for –, +, ++, +++ and unmeasurable, respectively). (a) Samples collected from participants without diabetes (n = 180) and (b) samples collected from participants with diabetes (n = 360). (c,d) The correlation charts between plasma glucose concentrations and logarithmically transformed values of the tear glucose concentrations before excluding samples with occult blood (black, blue, yellow, red and green circles stand for –, +, ++, +++ and unmeasurable, respectively). (c) Samples collected from participants without diabetes (n = 180) and (d) samples collected from participants with diabetes (n = 360).
Table S1 | Exclusion criteria.
Table S2 | Kolmogorov–Smirnov test for validation of normal distribution.