Identification of novel sublingual parameters to analyze and diagnose microvascular dysfunction in sepsis: the NOSTRADAMUS study

This prospective, observational, cross-sectional study took place in the medical and operative ICUs of the University Hospital Münster. The study was performed in accordance with the Declaration of Helsinki and was approved by the competent ethics committee (amendment to 2016-073-f-S). Some of the participants were already included in previous studies on eGC damage in sepsis [18, 20].

After written informed consent was obtained from the patients or their legal representatives, 34 adult ICU patients with sepsis were enrolled non-consecutively after initial resuscitation. Sepsis was defined by the sepsis-3 criteria published by the ESICM-SCCM Sepsis Redefinitions Task Force [22]. Exclusion criteria were age < 18 years, pregnancy, or oral mucosal inflammation or injury, which could locally influence the sublingual microvasculature. Seventeen apparently healthy volunteers served as controls.

Demographic variables, routine chemistry tests, and physiological variables, including the Sequential Organ Failure Assessment (SOFA) score [22] and a contemporary version of the Charlson Comorbidity Index (CCI) [23], were obtained for each subject at the time of sublingual videomicroscopy (Table 1). Each videomicroscopy set consisted of two complete measurements (see below) which were averaged to account for spatial heterogeneity of the sublingual microvasculature.

Figure 1 provides an overview of the process of video acquisition, data analysis and post-processing. Bedside intravital microscopy was performed with a sidestream dark field (SDF) camera (CapiScope HVCS, KK Technology, Honiton, UK) to visualize the sublingual microvasculature according to a standardized procedure as described in detail previously [20]. A physician experienced in the SDF technique, trained to recognize and avoid pressure artifacts performed the image acquisition based on the current round table recommendations [6]. The SDF camera uses green light-emitting stroboscopic diodes (540 nm) to detect the hemoglobin of passing red blood cells (RBCs). Using a 5 × objective with a 0.2 numerical aperture, images were captured, providing a 325-fold magnification in 720 × 576 pixels at 23 frames per second as described in detail previously [19, 24, 25].

The analysis of all variables was carried out exclusively with the GlycoCheck™ Software (Microvascular Health Solutions Inc., Salt Lake City, UT, USA) software. Briefly, the GlycoCheck™ Software allows video acquisition after predefined image quality criteria (motion, intensity, and focus) are fulfilled. Specifically, light intensity range is automatically set to prevent under- or over-exposure of the video. The acceptable motion range is set so that every vessel can be tracked in all frames of each video. Each complete measurement consists of at least ten 2-s videos (40 frames/video), containing a total of about 3000 vascular segments of 10 μm length each. All videos are deliberately obtained from different positions to counterbalance spatial heterogeneity of the sublingual microcirculation. The software automatically subjects the obtained vascular segments to a strict quality check, as previously described in detail [25]. Briefly, vascular segments are considered valid only if: (1) red blood cell content is ≥ 50%, (2) RBC width is ≥ 2 µm, (3) segments are neither curved, (4) nor too close to each other. Invalid vascular segments are marked yellow and are automatically discarded, while all valid vascular segments (green lines) are further analyzed. Therefore, all further parameters are calculated based on these valid segments. Screenshots of the sublingual mucosa of five randomly chosen individuals are shown in Additional file 1: Fig. S1 (with and without automatically acquired vessel detection and quality check). Moreover, two additional movie files illustrate the sublingual mucosa of a healthy control and a sepsis individual with and without automatic vessel detection and subsequent quality check in more detail (Additional file 2: Video 1, Additional file 3: Video 2). Finally, the software obtains up to 840 radial intensity profiles for each valid vascular segment based on the RBC column width (RBCW), and automatically groups vessels from 4 to 25 µm diameter in 22 separate diameter classes (1 µm each). This concept has been successfully used and validated in the past [25]. Data from two complete measurements (hereafter referred to as “measurement set”) were extracted, analyzed, and averaged offline to avoid sampling error and to counterbalance spatiotemporal heterogeneity of the sublingual microcirculation [19, 20].

Data are presented as absolute numbers, percentages, or medians with corresponding 25th and 75th percentiles (interquartile range; IQR), as appropriate. The nonparametric Mann–Whitney U test and the chi-square test were used to compare variables between patients and controls. To correct for multiple testing in comparisons of microcirculation variables per-diameter class we used the false discovery rate (FDR) approach of Benjamini, Krieger and Yekutieli, setting a qvalue < 0.05 as significant. Spearman rank correlation coefficient was used to assess correlations between clinical and microvascular variables. Associations between microvascular measures were evaluated using simple and adjusted linear regression models. All the tests used were two-sided, and statistical significance was set at p < 0.05. Our study was powered to detect a moderate correlation (Spearman correlation coefficient = 0.5) between microvascular health score and SOFA score in the septic cohort with 85% power given a two-sided alpha of 0.05 [27]. SPSS version 26 (IBM Corporation, Armonk, NY, USA) and GraphPad Prism version 8.4.3 (GraphPad Prism Software Inc., San Diego, CA, USA) were used for statistical analyses and preparation of figures.

First, we compared vascular density, V_RBC and PBR_static between healthy controls and sepsis patients in a diameter-class-wise fashion. This approach revealed a statistically significant decrease in vascular density only in the diameter classes 5, 6 and 7 µm in sepsis patients (− 63%, − 42%, and − 28% compared to controls), whereas the remaining diameter classes from 8 to 25 µm were not different between the groups (Fig. 2a and Additional file 8: Fig. S5). However, vascular density was not different between the groups in a pooled analysis (4–25 µm range) (12.3 vs. 12.8 mm/mm², p = 0.34). To check our results for plausibility, we correlated, in an explorative manner, vascular density with clinical variables, such as interleukin-6 (IL-6), procalcitonin (PCT), lactate and SOFA score. We found a robust inverse correlation in the 4 to 7 µm diameter range, while these associations were absent or even partially reversed in larger vessels (Fig. 2b and Additional file 7: Table A2).

V_RBC showed a huge overlap between patients and controls. Median V_RBC trended to be slightly lower in sepsis patients, especially in smaller vessels and was inversely correlated with variables of disease severity (Fig. 2c, d and Additional file 7: Table A2). A significant sepsis-induced increase in PBR_static was particularly apparent in feed vessels, probably due to loss of the affected smaller vessels (lower capillary density) (Fig. 2e and Additional file 9: Fig. S6). PBR_static values of individual diameter classes showed weak-to-moderate positive correlations with variables of sepsis severity (Fig. 2f and Additional file 7: Table A2).

Taken together, the diameter class-wise data analysis revealed individual sepsis-induced perturbations in vascular density, V_RBC and PBR_static.

Having observed a significant decrease in vascular density in very small capillaries (diameter (D) ≤ 7 µm) in sepsis patients, we determined and focused on the capillary blood volume (Fig. 3). CBV_absolute decreased from 16.5 [10.3–19.4] 10³μm³ in healthy controls to 7.9 [5.9–14.5] 10³μm³ in sepsis patients (p = 0.006). To further improve discrimination between the groups, we added some functional variables to the CBV calculation.

First, we calculated CBV_static, by multiplying CBV_absolute with a ratio derived from V_RBC in feed vessels relative to capillaries (V_RBC (D ≥ 10 µm)/V_RBC (D ≤ 7 µm)). However, the suspected decrease in the CBV ratio (due to decrease in capillary density) in sepsis patients did not occur (Fig. 3a). The reason for this only became apparent after plotting capillary V_RBC (D ≤ 7 µm) as a function of large vessel V_RBC (D ≥ 10 µm) (Fig. 3b). This type of analysis revealed a strong dependency between V_RBC (D ≤ 7 µm) and V_RBC (D ≥ 10 µm) in sepsis patients (R² = 0.63, p < 0.0001), indicating impaired capillary (de-)recruitment in this group. In contrast, capillary V_RBC is relatively constant in healthy controls as reported before [28], indicating functioning (de-)recruitment of CBV associated with changes of feed vessel blood flow in healthy subjects. CBV_static decreased from 18.5 [11.1–22.3] 10³μm³ in healthy controls to 10.2 [7.1–17.1] 10³μm³ in sepsis patients (p = 0.005).

Next, we calculated capillary recruitment (CR) as 1- the slope(V_RBC (D ≤ 7 µm), V_RBC (D ≥ 10 µm)). CR per group decreased from 78% in healthy controls to 8% in sepsis patients (Fig. 3c). Finally, we calculated CBV_dynamic by multiplying CBV_static with (1 + CR). CBV_dynamic showed a significant higher and much wider normal range in healthy controls compared to CBV_absolute and CBV_static, respectively. CBV_dynamic decreased from 32.8 [19.7–39.5] 10³μm³ in healthy controls to 11.1 [7.7–18.5] 10³μm³ in sepsis patients (p < 0.0001). Overall, we were able to show that by adding dynamic variables to CBV calculation, the discrimination between the groups could be significantly improved (Fig. 3d).

We have previously shown that PBR values (averaged across all diameter classes from 5 to 25 μm) increase in sepsis patients [18‐20], indicating sepsis-induced damage of the eGC. In this study, PBR_static (D 4 to 25 µm) increased from 2.24 [2.13–2.35] µm in healthy controls to 2.48 [2.33–2.62] µm in sepsis patients (p < 0.0001) (Fig. 4a). In sepsis patients, PBR_static showed a tendency to increase at low levels of feed vessel velocity, whereas in healthy controls, PBR_static showed a tendency to stay constant or even slightly decrease at low feed vessel velocity (Fig. 4b). To eliminate the influence of different RBC velocities on PBR_static, we estimated PBR_dynamic at a feed vessels V_RBC of 0 µm/s based on the different slopes(PBR_static, V_RBC (D ≥ 10 µm)) in healthy subjects and sepsis patients (Fig. 4b and Additional file 6: Fig. S4). PBR_dynamic increased from 1.95 [1.84–2.08] µm in healthy controls to 2.58 [2.43–2.76] µm in sepsis patients (p < 0.0001) (Fig. 4c). In summary, adjustment of the PBR to V_RBC improved discrimination between the groups by about 50%.

We have previously shown, that glycocalyx damage and microcirculatory impairment do neither coincide, nor do they occur in proportion in every sepsis patient [20]. This un-coupling of changes in PBR and CBV can also be reproduced in the group of sepsis patients (Fig. 5a–c). To account for this finding, we combined the different variables into one Microvascular Health Score (MVHS™) (Additional file 10: Fig. S7). Because CBV decreases and PBR increases in sepsis, we formed the quotient of CBV/PBR to calculate the MVHS in a static and a dynamic version (MVHS_static = CBV_static (D 4 to 6 μm)/PBR_static (D 4 to 25 μm) and MVHS_dynamic = CBV_dynamic (D 4 to 6 μm)/PBR_dynamic (D 4 to 25 μm)).

The median MVHS_static was significantly lower in sepsis patients 1.70 [1.32–2.53] points compared to healthy controls 4.03 [2.11–4.33] points (p < 0.001). The median MVHS_dynamic stressed the difference between sepsis patients and healthy controls (1.78 [1.38–2.67] vs. 7.43 [4.65–8.73] points, p < 0.0001) even more than the MVHS_static.

Both, MVHS_static and MVHS_dynamic correlated moderate-to-strong with SOFA score, number of dysfunctional organs, lactate, CRP, IL-6 and PCT (all p < 0.001) in a pooled analysis that included healthy controls and sepsis patients (Table 2). These observations remained unchanged in additional sensitivity analyses (Additional file 7: Table A3).

In the subgroup of sepsis patients, the correlations of MVHS_static and MVHS_dynamic with SOFA score and number of dysfunctional organs remained significant (all p ≤ 0.01) (Table 2 and Additional file 7: Table A4). MVHS_dynamic was significantly different in sepsis subgroups stratified by either median SOFA score or median number of dysfunctional organs, respectively (Fig. 5d and Additional file 11: Fig. S8).

Observing that RBC dynamics differ between capillaries (D ≤ 7 µm) and feeding vessels (D ≥ 10 µm) led us to relate capillaries and feeding vessels to each other—which resulted in the calculation of capillary recruitment. The concept of capillary recruitment in skeletal muscle was first proposed by August Krogh in 1919 [29]. Krogh hypothesized that the opening of previously closed muscle capillaries, i.e., capillary recruitment, would allow capillary RBC velocities to remain low, and capillary oxygen extraction thereby efficient, despite large increases in blood supply. While the binary distinction between "closed" and "open" capillaries, was key to Krogh's model argument, modern observations suggest that capillary recruitment should be regarded rather in the context of continuous than binary changes in RBC distributions and velocities among capillaries [30, 31]. In this respect, our quantitative data show that healthy individuals (at least under resting conditions) have relatively constant median capillary RBC velocities, which seems to be independent of the associated V_RBC in the feed vessels. Even though the underlying (auto-) regulation mechanisms remain incompletely understood, compliance with a narrow RBC velocity range seems plausible to guarantee an optimal supply–demand ratio in capillaries. Conceptually, constant capillary RBC velocities in the face of varying V_RBC in larger feeding vessels is consistent with a) increasing numbers of RBC-perfused capillaries at tissue sites with higher metabolic demand and increased blood supply in feeding vessels, as well as b) reduced numbers of RBC-perfused capillaries at sites with lower metabolic demand and reduced feeding vessel blood supply. The failure to maintain constant RBC velocities in capillaries of patients with sepsis and the fact that capillary RBC velocities in our septic cohort change proportionally with RBC velocity changes in feedings vessels, reflects that the number of perfused capillaries in sepsis patients is fixed and insensitive to local variations in tissue metabolic demand. A simplified version of the pathophysiologic concept of capillary (de-)recruitment observed in this study is visualized in Fig. 6.

Estimates of capillary recruitment in our cohort are based on per group analysis of relations between RBC velocities in capillaries and feeding vessels. New recording and analysis strategies that allow for calculation of capillary recruitment on a per-patient basis are currently under investigation. This modification will allow for capturing enough low- and high flow situations per subject to generate intra-individual regression slopes. Thus, it will be possible to determine the capillary recruitment per individual subject, which might further improve discrimination of dynamic variables, such as CBV_dynamic and MVHS_dynamic.

The sepsis-associated PBR increase measured in this study is consistent with changes observed in previous studies [18‐20, 32‐34]. The current study shows that adjustment of PBR estimates to V_RBC improved discrimination between the groups by about 50%. This gain in discriminatory power is important, as PBR estimates have so far shown a considerable overlap between healthy controls and sepsis patients. This adjustment accounts for RBC velocity dependency of the glycocalyx [35, 36]. Especially in sepsis patients, PBR tends to increase when RBC velocity decreases, suggesting that glycocalyx extends towards a more porous and permeable compartment in the absence of exposure to fluid/RBC shear forces. The fact that PBR increases are more extreme in the sepsis patients, is consistent with previous studies showing that glycocalyx is damaged and more permeable in sepsis, which likely contributes to its increased RBC velocity dependency. However, it is difficult to estimate if the higher PBR observed in sepsis patients is caused solely by real glycocalyx damage (i.e., enzymatic shedding [18, 37]) or if temporary and functional changes that enable nutrient and solute supply to the interstitial space contribute to that PBR increase. We have previously shown that PBR estimates correlate excellent with glycocalyx thickness measured by atomic-force microscopy in cultured endothelial cells exposed to concordant sera from healthy subjects and sepsis patients [18, 20]. Glycocalyx damage in vitro was completely abolished when enzymatic activity of heparanase, a heparan sulphate specific endo-beta-D-glucuronidase, was blocked by addition of heparin [18]. Furthermore, circulating levels of Syndecan-1, a core protein of the intact glycocalyx, increased tenfold in sepsis patients and correlated with PBR values [20], indicating that enzymatic shedding is a dominant mechanism of PBR increase in sepsis. Elegant work in the cremasteric microcirculation showed that, inflammation-induced glycocalyx shedding increases effective capillary diameter. This effect is, however, defeated by enhanced white blood cell (WBC)-endothelial interactions and subsequent venular WBC obstruction, resulting in a net reduction of capillary V_RBC [38]. Glycocalyx damage may therefore partly explain the trend towards lower V_RBC in capillaries.

In the current cross-sectional cohort, the proposed MVHS_static and MVHS_dynamic did correlate with SOFA score and number of dysfunctional organs in the subgroup of sepsis patients. Correlations between classical variables of the microcirculation and the SOFA score have been rarely reported. The SOFA correlated negatively with microvascular flow indices (MFI) as well as total and perfused vascular densities (TVD and PVD) in sepsis patients receiving activated protein C [33]. However, we couldn´t detect any significant correlations of MFI, TVD or PVD with the SOFA score in a previous sepsis study [20]. The finding that the MVHS performed better than its individual components is a strong argument for combining different measures of microvascular dysfunction such as vascular blood volume (CBV) and glycocalyx properties (PBR) into one score. In our view, in particular the combination of CBV and PBR makes sense, because CBV and PBR appear to behave independently of each other [20], and may thus indicate different aspects of microvascular dysfunction. To clarify whether the MVHS can indeed predict relevant outcome in sepsis patients, we initiated two prospective, observational, longitudinal studies to evaluate the MVHS in the emergency room (Early Detection of Glycocalyx Damage in Emergency Room Patients—the EDGE Study, Clinicaltrials.gov Identifier: NCT03126032) and in the ICU (Analysis of Sublingual Glycocalyx Damage at ICU Admission to Predict Risk of Death—the ASGARD Study, Clinicaltrials.gov Identifier: NCT03847493). In these studies, serial measurements can possibly reveal differential changes in individual components of the MVHS in response to the initial therapy.

We acknowledge some limitations of our study. First, it is a single-center study with a limited sample size. Therefore, the findings cannot be directly generalized for all sepsis patients. However, the interdependencies between our novel variables and their associations with clinical variables remained significant after multiple adjustments and additional sensitivity analyses. Second, this study was neither designed nor powered to test the performance of novel variables for outcome prediction. However, our findings add interesting new aspects to the lively field of intravital microscopy research. Further longitudinal studies are needed to evaluate the MVHS for outcome prediction. Third, since all calculations are ultimately based on the flow properties of RBCs, we can only analyze microvessels in which a minimal number of RBCs are present and the predefined quality criteria are met. Vessels without RBCs or invalid vascular segments are therefore not detected by the software. That could have an impact not only on the density calculations, but also on the estimation of the PBR, as severely affected capillaries might not be accessible to RBCs anymore. However, past studies have shown, that the average PBR (4–25 µm) of healthy and septic subjects measured in vivo correlates excellently with the decrease in glycocalyx thickness of endothelial cells after incubation with the concordant serum samples, further supporting the accuracy of PBR values. Fourth, technical limitations, as, e.g., pressure artifacts or low spatial resolution might have affected the results. However, all measurements were performed by only one very experienced investigator specially trained to avoid pressure artifacts. Moreover, the camera has a sufficiently high resolution.