A Caenorhabditis elegans behavioral assay distinguishes early stage prostate cancer patient urine from controls

ABSTRACT Current methods for non-invasive prostate cancer (PrCa) detection have a high false-positive rate and often result in unnecessary biopsies. Previous work has suggested that urinary volatile organic compound (VOC) biomarkers may be able to distinguish PrCa cases from benign disease. The behavior of the nematode Caenorhabditis elegans has been proposed as a tool to take advantage of these potential VOC profiles. To test the ability of C. elegans Bristol N2 to distinguish PrCa cases from controls, we performed chemotaxis assays using human urine samples collected from men screened for PrCa. Behavioral response of nematodes towards diluted urine from PrCa cases was compared to response to samples from cancer-free controls. Overall, we observed a significant attraction of young adult-stage C. elegans nematodes to 1:100 diluted urine from confirmed PrCa cases and repulsion of C. elegans to urine from controls. When C. elegans chemotaxis index was considered alongside prostate-specific antigen levels for distinguishing cancer from cancer-free controls, the accuracy of patient classification was 81%. We also observed behavioral attraction of C. elegans to two previously reported VOCs to be increased in PrCa patient urine. We conclude nematode behavior distinguishes PrCa case urine from controls in a dilution-dependent manner.


INTRODUCTION
Many cancers have altered metabolic pathways that support, or may even cause, malignancy [1].
Detection of these cancer-related metabolites or volatile compounds in blood and urine would provide an ideal means of non-invasive early cancer detection. Yet, brute-force chemical analyses have been unsuccessful in determining metabolite profiles that consistently distinguish early cancer from healthy biofluids [2]. Remarkably, even in the early stages of malignancy, there is evidence that cancer patients emit odors that can be accurately detected by canine and murine olfaction [3][4][5][6][7]. Replicating this process of scent identification in mammals using technology or computation is hindered by the exceptionally complex neural processing in mammals that occurs to identify unique odors [8]. Further, mammalian detection systems rely on training and learned memory to discriminate between odors, and efficacy may be affected by animal personality, genetics, or environment [7]. Therefore, animal systems that have a naturally occurring odor discrimination for cancer patient samples could be a powerful tool for developing early cancer detection technology.
Recent work by Hirotsu et al. [9] demonstrated the feasibility of a "Nematode Scent Detection Test" (NSDT) to take advantage of the chemosensory abilities of Caenorhabditis elegans, a small nematode worm. C. elegans is commonly utilized as an experimental model in neurobiology because of its simple nervous system that is primarily devoted to chemosensation [10]. In addition, with a sizable fraction of its genome being dedicated specifically to olfactory components, the availability of genomic tools makes C. elegans an appealing system for understanding sensory mechanisms [11]. C. elegans has a well-documented ability to detect a wide variety of volatile and water-soluble compounds necessary to differentiate food, mates, pathogens and predators encountered in its natural environment [10,12]. Using a simple assay setup, Hirotsu et al. [9] demonstrated that C. elegans can detect multiple types of cancer, and potentially even pre-cancer, from 1:10 diluted urine with >95% specificity and sensitivity. This assay, known as a chemotaxis assay, is a commonly performed lab assay and has been used to score the relative attractiveness or repulsiveness of various volatile organic compounds (VOCs) to live C. elegans nematodes [12][13][14]. Hirotsu et al. [9] further showed the C. elegans behavioral response worked strictly through Gprotein coupled receptor (GPCR) mediated signaling and via chemosensory neurons known to be C. elegans N2 nematodes demonstrate an overall attraction to PrCa patient urine compared to controls.
All chemotaxis assays had at least 6 replicates performed except for individuals N03, B04, and C02, which had 4. The average number of worms that chemotaxed in each assay for negative screen, benign, and cancer groups was 37.4 (s.e.m. = 1.7), 38 (s.e.m. = 1.7), and 34 (s.e.m. = 1.3), respectively. We observed both attractive and repulsive patient urine samples within the three groups (Fig. 2). After CIs for each patient were averaged, urine from PrCa patients elicited generally positive average CIs compared to urine from benign and negative screen individuals (Fig.   3). Shapiro-Wilk normality test results indicated the distribution of measured CIs did not significantly differ from a normal distribution for negative screen (P = 0.45), benign (P = 0.91), or cancer (P = 0.95) groups. One-way statistical analysis revealed a significant difference in means among the three groups (one-way ANOVA, Fdf=2 = 4.71, P = 0.012). A Tukey's HSD post hoc test indicated that average CIs for PrCa were significantly higher than those of benign (P = 0.020) and negative screen (P = 0.034), while benign and negative screen types were not significantly different (P = 0.896). We found that PrCa urine samples CIs were significantly higher compared to benign and negative screen individuals when all technical replicates (i.e. no subsampling) were considered for each patient sample (one-way ANOVA, Fdf=2 = 4.71, P = 0.0045, Fig. S2).

C. elegans N2 CI values do not significantly associate with patient clinical profile features.
We assessed potential pairwise associations between average CIs and PSA, BMI, age at collection, and prostate size. C. elegans N2 CIs did not correlate significantly with PSA, BMI, or age (Fig. 4).
We also found no difference in average CIs between current/former smokers (n = 25) and nonsmokers (n = 41) (Welch Two-Sample t-test, P = 0.81; Supplemental Fig. S3, left). Similarly, we found no significant difference in the average CIs of low (3+3 or 3+4, n = 13) and high (>4+3, n = 8) Gleason score tumors (Welch Two-Sample t-test, P = 0.23) within the PrCa group (Supplemental To calculate sensitivity and specificity, we used classification as determined by pathology. True positives were considered to be individuals with a confirmed case of PrCa following biopsy, while true negatives were considered to be individuals that either had a negative biopsy or were deemed to not have PrCa following screening. To test the ability of average C. elegans N2 CIs to predict cancer status, we used a model using CI > 0 to classify cancer and CI < 0 to classify cancer-free. With this CI model, a sensitivity of 76% and a specificity of 67% was determined for discriminating slightly lower than the 71% accuracy of PSA alone (balanced accuracy = 77%). Neither CI nor PSA alone were determined to have a significantly higher accuracy than the no-information rate (NIR). In addition, both CI and PSA classification models alone were determined to have significant Mcnemar's Test P-values, and therefore both models are presumed to be fundamentally different from the classifications made by pathological assessment.
We next tested a combined classification model that required both CI > 0 and PSA ≥ 4 ng/ml for a PrCa diagnosis and compared the model to using CI or PSA alone. Using the combined model, we were able to increase the specificity to 85% at the cost of reducing the sensitivity to 71%. The combined model accuracy improved to 81% (balanced accuracy 78%). In addition, the combined model had significantly better accuracy than the NIR of 66% (P = 9e-3). Classification model performance comparisons are summarized in Table 2.

DISCUSSION
Accumulating evidence suggests that the metabolomic profiles of patients with prostate cancer can be distinguished from normal patients and this difference could potentially be measured through blood or urine samples [20]. The leading metabolite candidate for prostate cancer, sarcosine, has remained controversial as a specific urinary biomarker and has yet to find widespread adoption in the clinic [21][22][23]. Thus, the search for alternative methods of cancer metabolite biomarker discovery continues to be of great interest.
The potential for animals to sense malignancy has received increasing attention since the phenomenon was first described over three decades ago [24]. Since the first report of C. elegans accurately classifying cancer samples by Hirotsu et al. [9], similar methods have been applied using C. elegans to detect sepsis [25] and tuberculosis-specific odorants [26]. Two of the VOCs proposed to be increased in PrCa urine samples by Khalid et. al [29] were found to elicit behavioral attraction of C. elegans under our assay conditions. When chemotaxis assays were performed using urine samples from our cohort, our overall findings agree with the previous work by Hirotsu et al. [9] and support a dilution-dependent behavioral response of C. elegans nematodes to cancer patient and control urine. It should also be noted that only one identified case of PrCa was tested in the Hirotsu et al. [9] cohort, and thus direct comparisons of results are difficult. Interestingly, some urine dilutions elicited opposite behavioral responses (i.e. switching from attractive to repulsive) compared to other dilutions of the same sample. This finding was also observed by Hirotsu et al. [9] and may be due to the complex VOC composition of human urine [27] causing concentrationdependent preference changes as dilutions change [14]. Prior work by Cornu et al. [28] reported 91% sensitivity and specificity using canine olfaction trained exclusively on urine samples from prostate cancer patients and controls. The Cornu et al. [28] sample cohort included 33 urine samples from confirmed stages I-IV prostate cancer cases and 33 control samples from patients with negative biopsies. Our results also suggested significant olfactory discrimination of C. elegans Biology Open • Accepted manuscript to prostate cancer patient urine compared to control patients, although our observed sensitivity and specificity was lower than the results reported by Cornu et al. [28]. This reduced performance may be due in part to our cohort consisting of only early stage (stages I-II) prostate cancer samples. It is possible that there is an increase in the amount or composition of attractive VOCs as prostate cancer progresses [20]. Our data do not indicate a clear preference by C. elegans for low (3+3 or 3+4) or high (>4+3) Gleason score tumors, and the overall effect of tumor stage on animal olfaction remains to be determined.
Interestingly, our accuracy of 66% using the C. elegans CI for patient classification is within the range of the ~66% accuracy reported by Khalid et. al [29] using four VOCs. Whether any of those four VOCs are the same ones the nematodes detect in the PrCa urine samples of our cohort is unknown. Importantly, it is possible that the utility of VOCs alone in PrCa urine sample classification is limited as a biomarker. This limitation could be due to tumor heterogeneity, wherein only some prostate adenocarcinomas produce the identifying VOC(s), or high among-individual variation in VOC expression that can mask signal. Importantly, however, our data suggest that the C. elegans CI in response to urine samples is independent of blood-derived PSA and combining the two measurements increased the accuracy of the classification model to 81%. A similar finding was reported by Khalid et. al [29] and later by Gao et al. [30] whereby the sensitivity and specificity of PrCa VOC profiling by GC-MS was significantly improved by also utilizing PSA. Thus, our study further supports supplementing PSA models with VOC profiles as a means to increase the accuracy of PrCa detection.
Because knowledge of the specific ligand-GPCR interactions that determine C. elegans chemotactic behavior remains limited [10,31,32], the exact olfactory receptors that may be responsible for C. elegans behavior-based cancer detection are unknown. Regardless of this limitation, there are approaches that could be further developed to leverage C. elegans olfaction as a diagnostic or biomarker discovery tool. The first approach is to develop high-throughput and reproducible technology that can use the animal behavior preference itself as a diagnostic assay (e.g., the "N-nose" demonstrated by Hirotsu et al. [9] and Kusumoto et al. [15]). Another possibility is to use the behavioral assay as an unbiased sensor system that can be coupled with a discovery method such as GC-MS. In this method, the behavioral assay has a similar function to an "electronic nose" but with a potentially increased limit of detection and no need for machine learning [33]. Biomarkers that can be obtained noninvasively and boost the accuracy of PSA for PrCa detection are highly desired. In this work, C. elegans demonstrated a weak but significant attraction to urine from PrCa patients when measured by behavioral assay. The C. elegans behavioral assay did not misclassify the same patients as PSA and combining the two outcomes increased overall accuracy and specificity. The potentially independent value of CI, and thus potentially of VOCs of interest, was further supported by the lack of CI correlation with other clinical attributes. While we were unable to determine the VOCs that elicited the behavioral response, our results support previous work that animal olfactory responses could be a useful tool for cancer biomarker discovery.

All experimental protocols were reviewed and approved by the Oregon Health & Science University
Institutional Review Board (protocol #18048). Informed consent was obtained from all patients included in this study. Urine from human subjects was acquired based on the following inclusion criteria: (1) male patients between 45-75 years old at time of consent, (2) no current or previous cancer diagnosis (excluding squamous cell/basal cell carcinoma), (3) a total PSA between 2.5-20 ng/ml drawn within the last two years, (4) no more than one previous negative biopsy, (5 (Table 1).

Nematode strains and culture conditions.
C. elegans Bristol N2 (referred to as "N2") was used for all chemotaxis assays. C. elegans N2, a commonly used lab strain, has experienced many generations of lab culture before being cryogenically stored as separate strain in 1980 (WormBase). Following Hirotsu et al. (2015), nematodes were maintained at 20°C on Nematode Growth Medium Lite (NGML; US Biological #N1005) plates seeded with a lawn of NA22 Escherichia coli as a food source.

Nematode age synchronization.
Prior to all experiments, strains were allowed to recover from freezing for two to three generations prior to use. In accordance with Hirotsu et al. (2015), young adult stage nematodes were used for all chemotaxis assays. All centrifugation steps were performed at room temperature (i.e., 21 o C).
Mixed-age nematode populations were age synchronized using standard bleach methods [37]. The resulting egg pellet was then resuspended into the 1 ml of remaining buffer solution and the eggbuffer solution was pipetted onto new NGML plates seeded with NA22 E. coli. The plated eggs were maintained at 20ºC for 72 hours to allow age-synchronized nematodes to hatch and reach the young adult stage.

Chemotaxis assays and calculation of chemotaxis index.
We used a chemotaxis assay plate format with diagonally opposing sample quadrants to reduce the potential of random worm movements affecting results. Following Margie et al. [13], unseeded NGML plates prepared without antibiotic were divided into 4 equally sized quadrants (Fig. 6). A circle with a diameter of 1 cm was drawn around the center of the plate. Quadrants across from one another were labelled with "T" for the sample (previously reported VOC or diluted urine) and the other two quadrants were labelled "C" for control diluent (water). Each quadrant contained a dot 2.5 cm from the center of the plate for placement of 1 µl of sample or water. Approximately 50 washed, young adult nematodes were plated in the center of the assay plate. We controlled for light by using clean, unscented towels to cover the chemotaxis plates during assays. The plates were left on the benchtop, covered with towels for 1 hr at room temperature (21°C). Assays were performed for 1 hr at room temperature to match the length and temperature conditions of the chemotaxis assays performed by Hirotsu et al. [9]. After 1 hr, the plates were individually photographed and the number of nematodes in each quadrant was recorded. Each completed chemotaxis assay was considered a technical replicate for statistical analysis. An example completed chemotaxis assay is shown in Supplemental Fig. S1.

Biology Open • Accepted manuscript
To calculate an overall score of behavioral response to a sample, hereafter referred to as the chemotaxis index (CI), the equation: CI = ((T1 + T2) -(C1 + C2)) / ((T1 + T2 + C1 + C2)), (Eqn 1) was used where T1 + T2 was the number of nematodes that migrated into the first and second quadrants containing the sample being tested and C1 + C2 is the number of nematodes that migrated into the first and second quadrants containing only water, respectively [13]. To control for nematodes that were immobile or incapable of olfaction, all animals in the 1 cm circle near the center were not counted. Thus, a positive CI when calculated using Eqn 1 indicates nematode attraction towards a sample, a negative CI indicates repulsion from a sample, and a CI = 0 indicates no behavioral preference. We define a technical replicate as a single completed chemotaxis assay using one diluted sample. At least 4-6 technical replicates were performed for each urine sample or VOC compound.

Chemotaxis assays on patient urine samples.
Prior to use in chemotaxis assays, urine samples were thawed at room temperature and inverted 3 times before being freshly diluted in sterile Milli-Q filtered water. The same diluted sample was used for all replicates performed on the same day. We first determined the optimal urine dilution that can distinguish cancer from controls by performing chemotaxis assays on serially diluted urine at 1:10, 1:50, 1:100, 1:500, and 1:1000 dilutions. For these assays, the same 4 patients per group were used for all 5 dilutions. We considered the best urine dilution to have 1.) the highest overall accuracy rate for discriminating cancer versus control samples, based on a positive or negative CI and 2.) the highest mean difference between CI when comparing cancer and control groups.
Following determination of the best urine dilution, we performed chemotaxis assays for the remaining urine samples only at that dilution. Each round of chemotaxis assays included 6 samples assayed at the same time: 1 unblinded sample from each of the three groups (negative screen, benign, and cancer) and 3 samples with blinded disease status. The disease status of the blinded samples was revealed following completion of all assays. In some cases, more than 6 chemotaxis assay technical replicates were performed in total for some unblinded samples. To prevent skewed oversampling of these individuals, 6 CI technical replicates were randomly sampled and those 6 CIs were used to calculate the average CIs for the individuals. Therefore, 4-6 chemotaxis assay CIs are averaged to calculate one CI per patient sample (i.e. biological replicate). As a positive control, at Biology Open • Accepted manuscript least 3 chemotaxis assay replicates using the known attractant isoamyl alcohol (TCI # I0289, >99.0% purity) as a sample were performed alongside urine sample assays for each assay block [12]. The isoamyl alcohol was diluted to 9.09 mM prior to use.

Data analysis.
Statistical analysis, subsampling, and plot generation was conducted using R (v.3.6.1) and Rstudio  [39]. P-values were adjusted for multiple comparisons and statistical significance was determined at P < 0.05 for all tests. Plots were generated using the "ggplot2" R package [40] For boxplots, potential outliers were determined using the default parameters in the "ggplot2" package. Figure 1. C. elegans N2 are attracted to two VOCs that were previously reported to be increased in prostate cancer patient urine compared to healthy controls. Chemotaxis assays were performed using 2-octonone (left) or pentanal (right) diluted with water. n = 6 chemotaxis assays per concentration; error bars represent mean ± s.e.m.  b ns, not significant; *, P < 0.05; **, P < 0.01; ***, P < 0.001. Figure S2. When all technical assay replicates are considered for each sample, C. elegans N2 are more attracted to urine from PrCa patients than to benign or negative screen patient urine. Chemotaxis assay technical replicates (n = 4-30 per individual patient) were averaged for negative screen (n = 27 individuals), benign (n = 19 individuals), and cancer (n = 21 individuals) patient urine samples prior to plotting and statistical testing. A Tukey's HSD post hoc test indicated that average CIs for PrCa were significantly higher than those of benign (P = 0.015) and negative screen (P = 0.0083), while benign and negative screen types were not significantly different (P = 0.999). Boxplots for each group are inset within their respective violin plots. *, p < 0.05; **, p < 0.01; ns, not significant. Figure S3. C. elegans N2 do not show differences in attraction to urine samples depending on donor smoking status or Gleason score. There was no significant difference in attraction when individuals without a history of smoking were compared to current or former smokers (left panel).

Figures
Similarly, there was no significant difference in attraction to urine from patients with tumors confirmed to have low (3+3 or 3+4) or high (>4+3) Gleason scores (right panel).