Over two weeks, the perfusion bioreactor system and imaging platform described here demonstrated the ability to obtain metabolic redox ratio data from 3D tissue engineered human adipose tissue samples in vitro
. The average cellular redox ratios gathered through TPEF imaging and subsequent image processing showed similar trends as has been observed in static cultures; consistent redox ratios over the first week were followed by a significant drop at 2 wks. The decrease in redox ratio from 1 to 2 wks is consistent with previous work and indicates hASC differentiation 
. Upon differentiation, hASCs are not expected to continue to proliferate. This expectation is supported in by the relationship between the average cellular redox ratios () and the average cellularity (). There is no significant change in redox ratio observed within the first week of testing, which would suggest that hASC differentiation is low and proliferation is continuing. This is further supported by the increase in cellularity observed from the 0 to 1 wk time points. Then as the rate of differentiation begins to increase between the 1 and 2 wk time points, the redox ratio decreases and cellular proliferation halts, as observed in the more modest increase in cellularity from 1 to 2 wks compared to the rapid increase from 0 to 1 wks.
Previous analyses demonstrated a decrease in redox ratio coincided with the formation of lipid droplets 
. The underlying mechanism driving the relationship between redox ratio and lipogenesis has been attributed to an accumulation of mitochondrial NADH during de novo fatty acid synthesis. Energy storage through fatty acid synthesis and lipid droplet formation likely only occurs when the relative energy consumption is low. Decreases in ATP production and NADH oxidation by the electron transport chain combined with an increase in the conversion of NAD+ to NADH by the pyruvate dehydrogenase complex when breaking down pyruvate for fatty acid synthesis collectively cause a decrease in the mitochondrial ratio of NAD+/NADH. Furthermore, FAD bound to lipoamide dehydrogenase in the pyruvate dehydrogenase complex will likely remain in the oxidized form when the less NAD+ is present to accept the electrons. Collectively, these mechanisms are thought to provide a consistent decrease in redox ratio during lipogenesis in both static and perfusion cultures.
The smaller decrease in redox ratio in the perfusion culture relative to static cultures could be a result of a slight delay or reduction in the number of hASCs differentiating and undergoing lipogenesis at 2 wks. Alternatively, the endothelial cells may be more viable or proliferative in the perfusion culture than in static cultures causing a dampening of the expected decrease in redox ratio associated with adipogenic differentation. However, previous comparisons between adipogenic differentiation in hASC monocultures and hASC-endothelial co-cultures yielded no differences between the redox ratio of the two groups beyond Day 1, suggesting endothelial cells likely have little effect on redox ratio changes over time. Nonetheless, CD-31 staining for endothelial cells in histological sections can be performed in future work to gain more insight into the immediate behavior of endothelial cells in the 3D cultures with respect to cellular proliferation and angiogenesis. It is also worth noting that experimental differences between the current perfusion study and previous static culture study 
, including a human endothelial cell source and the different solvents used during silk scaffold processing, may cause a small difference in the temporal pattern of differentiation. Finally, it is worth considering that the mechanical environment produced by perfusion may inhibit adipogenic differentiation or lipogenesis to some degree or selectively detach mature adipocytes with large lipid stores.
Previous studies involving adipose tissue engineering and subsequent characterization for clinical applications have set forth both protocols and analytical benchmarks by which adipose tissue can be analyzed in vitro
and compared. Adipose tissue has been engineered using both synthetic and natural polymer scaffolds with both human mesenchymal and pluripotent stem cells 
. These studies have provided an understanding of how lipid droplet formation, cellular proliferation, lumen formation, and cell metabolism change over time. These observations demonstrate cellular distribution, ECM growth, and vascularity within engineered adipose tissues grown in static culture. Through these previous studies, a standard set of analytical tools used to evaluate the 3D engineered adipose tissue have been developed which include, but are not limited to, histology, immunohistochemistry, and two-photon imaging 
. In this study histological analysis with H&E staining was performed and is capable of measuring this cellular distribution and ECM growth, which is helpful in supporting and verifying the data collected from TPEF imaging. In previous work, the relationships have been established between TPEF imaging and Oil Red O staining for lipid droplet formation, as well as CD31 staining for endothelial cell organization 
. Future work with the bioreactor system used in the current study will further confirm that TPEF relationships with these other traditional analytical benchmarks are maintained in a perfusion environment.
As observed in , there is little, if any, ECM present in the 0 wk samples. This is likely because these samples had only been seeded with differentiating hASCs for two days, which is not sufficient time to accommodate ECM formation. There is, however, ECM present in the 1 and 2 wk cultures (note red arrows on ). This is likely a result of increased cellularity within the tissues, accomplished through more allowed time for the cells to adhere to the silk scaffolds and generate an ECM. The ECM of adipose tissue primarily consists of a loose collagenous matrix to provide structural support. It would be expected that if these tissues had remained in culture for longer times, that the cellularity of the tissues would increase while a more robust and expansive ECM developed. We have previously reported such findings using more traditional destructive modes of imaging for engineered 3D human adipose tissue systems. ECM and cellularity have been visually observed through immunohistochemistry, Oil Red O, and image processing cell count algorithms 
Although substantial variability was observed in cell densities, there did appear to be a trend toward higher cell density in the superficial layer of the tissue at 0 wks compared to the deeper sections. This is not surprising because the cells were only just seeded, and most likely did not have much opportunity to migrate throughout the silk scaffolds. Aggregation of the cells was observed within the superficial (0–40 µm) range at the 1 wk time point within the perfusion cultures. However, by 2 wks, cellularity appeared to level out within each of the layers (0–40 µm, 80–120 µm. 160–200 µm) in both the noninvasive imaging analysis () and H&E stains in the perfusion culture (). Longer experimental time points would be required to confirm this trend, but a more homogenous distribution of cells within a 3D scaffold in perfusion culture environment relative to static has been observed elsewhere 
Relative to normal tissue culture plate cell observations, the perfusion bioreactor and imaging system affords distinct advantages, including more physiologically relevant perfusion flow and the ability to perform repeated two-photon imaging analyses with submicron resolution. The impact of this approach is improved efficacy and relevance of future analytics by providing a platform for continuous metabolic and structural analysis of 3D human tissue samples. The approach also implicitly allows for improved cost and labor efficiency by offering an alternative to non-destructive analyses, saving substantial resources. Certainly, optical microscopy evaluation is limited to the outer 0.5–1 mm of the tissue. While no depth-dependent differences in redox ratio were observed in this or previous studies 
, additional evaluation of unstained tissue sections may provide further insight into the necessary scaffold and bioreactor design considerations for high-throughput drug and toxicity screening applications.
In conclusion, the research here demonstrates the ability to interface non-linear optical microscopy with 3D human tissue constructs within a perfusion environment for noninvasive continuous analysis. The perfusion bioreactor system was modified to be compatible with advanced imaging capabilities using high numerical aperture objectives, but was still capable of sustaining continuous flow over weeks to months with media replenishment. The perfusion bioreactor system sustained cultures over the span of weeks, and demonstrated potential to continue for months, as has been previously performed in static cultures. Multiple samples that entered the perfusion system were able to remain in a sterile environment and maintain viability throughout the experiment. TPEF imaging provided metabolic and cellular distribution information which correlated with standard histological analysis. These destructive and non-destructive imaging approaches supported each other, both showing adipose tissue formation as characterized by an increase in cellularity throughout testing, and ECM growth as shown through histology. A constant redox ratio was observed over the first week of testing in both the current perfusion study and previous static study 
, but both demonstrated a significant decrease in redox ratio from the 1 to 2 wk time points, suggesting lipogenesis. This combination of quantitative metabolic imaging and 3D tissue engineering using perfusion bioreactors represents a step toward low-cost, high-throughput methods to study more physiologically relevant engineered human tissue systems in applications such as disease models, drug screening, or developmental biology.