Alignment files (bam files) for 92 White and 19 Black LSCC patients were downloaded from The Cancer Genome Atlas Head-Neck Squamous Cell Carcinoma (TCGA-HNSC). All bam files were converted to fastq files using bedtools (29). The raw reads from fastq files were preprocessed using mapper.pl from miRDeep2 (30). Quality control was performed by removing reads with alphabets other than a, c, g, t, u, n, A, C, G, T, U, N and reads less than 15 nucleotides long from downstream analysis. The remaining reads were aligned to human miRNA precursors downloaded from miRBase release 21 (31) using quantifier.pl from miRDeep2. Alignment between precursor and mature miRNA was performed to generate the final miRNA counts. In doing this, mature miRNA sequences were first downloaded from miRbase and aligned to their miRNA precursors. Then, the alignment between mature miRNAs and the reads were compared and the number of reads falling within 2nt upstream and 5nt downstream of the corresponding miRNA was taken as the read counts for that miRNA.

Differential expression analysis was performed to compare miRNA expression between Black and White tumor samples. Counts were normalized using Relative Log Expression (RLE) implemented from edgeR (32). A negative binomial generalized log-linear model implemented in edgeR was used for differential analysis. EdgeR differential expression analysis was performed using the White tumor group as the reference by default; the direction of the fold-change was reversed post-hoc to consider changes in Black tumor samples relative to White tumor samples. Significantly differentially expressed miRNAs identified at a fold change >1.5 and a p value < 0.05.

Sex was not considered as a biological variable due to the limited sample size of females within the TCGA dataset used for analysis. Randomization of the TCGA cohort was irrelevant because the study was specifically designed to explore differential miRNA expression by race. Blinding was also deemed irrelevant to the study design. Power analysis was not conducted for the RNA-Seq TCGA patient data because our exploratory data analysis was constrained by the limited data points available within the database following race stratification (92 White versus 19 Black).

Human laryngeal squamous cell carcinoma cell lines, UM-SCC-12 (Black patient derived; RRID: CVCL_7717) and UM-SCC-10A (White patient derived; RRID: CVCL_7713) were authenticated via short tandem repeat typing and further genetically characterized (33, 34) prior to purchase from the University of Michigan Head and Neck cell line repository. Both cell lines were age, sex, grade and stage matched (see Supplementary Table 1 ). Cells were cultured in a T 75cm² flask containing Dulbecco Modification of Eagle’s Medium 1X (DMEM, 10-013-CV, Corning) supplemented with 10% heat inactivated fetal bovine serum (FBS, 35-011-CV, Corning), and 2% Penicillin/Streptomycin (PENSTREP, 15-140-122, Gibco) within a humidified incubator containing 5% CO₂ at 37°C. Cells were utilized in the following assays upon reaching 80% confluence.

In a 12-well format, UM-SCC-12 or UM-SCC-10A cells were reverse transfected with 50nM of either miR-9 mimic or inhibitor, respectively (see Supplementary Table 3 for oligo sequences and product information). Transfection efficiency was enhanced through the use of Lipofectamine RNAiMAX (13778075, Thermo Fisher) and Opti-MEM I Reduced Serum Medium (31985062, Thermo Fisher) per manufacturer’s instructions. Cell growth was optimized to ensure 60-80% confluency by assay endpoint. Length of transfection was dependent upon validation assay performed. All assays were performed in triplicate.

The scratch wound assay was used to assess cell migration. 24 hours post transfection, the confluent cell monolayer was disrupted using a 1000µL pipet tip. Images were captured at 0h, 24h, 48h, and 72h post cell monolayer disruption using the EVOS FL Cell Imaging System (ThermoFisher). The wound healing size tool plugin for ImageJ (RRID: SCR_003070) was used to quantify wound healing at each timepoint.

Baseline IC50 for cisplatin in UM-SCC-12 and UM-SCC-10A was determined with serial dilutions of Cisplatin (1134357, Millipore Sigma) at concentrations of 200µM, 66.67µM, 22.22µM, 7.41µM, 2.47µM, and 0µM. 48 hours post cisplatin treatment, cisplatin and spent media were aspirated from the wells and a 1:7 dilution of the Cell Titer Blue reagent (G8080, Promega) was applied to the cells in the wells. Baseline fluorescence (560/590 nm) was assessed using the BioTek SYNERGY H1 Multi-Mode Microplate Reader. The gain was adjusted such that all baseline values were similar ~2000nm. The plates were incubated in 5% CO₂ at 37°C with subsequent plate readings taken in 30 min intervals for 4 hours. IC50 to cisplatin was calculated in GraphPad Prism Version 9.40 (RRID: SCR_002798) using percentage of cell viability values. Percentage of cell viability was calculated as follows:

24 hours post transfection with a specific oligo and appropriate control, cells were washed with 1mL of 1x dPBS (1x dPBS, 21-031-CV, Corning) per well. Cells were harvested with 1x TrypLE (12604-013, Gibco), counted and replated as five replicates per condition in a 96-well plate. The next day cells were treated as described above at 48 hours cisplatin treatment.

Average fluorescence values generated at no treatment dose (0µM of cisplatin) in transfected cells corresponded with baseline cell proliferation. As such, final average fluorescence values at the 0µM dose were normalized against their baseline average fluorescence values to calculate relative fluorescence which represented cell proliferation in transfected cells.

24 hours post transfection, cells were washed with 1mL 1x dPBS per well and collected in cold TRIzol Reagent (15596018, Thermo Fisher). RNA was isolated per manufacturer’s instruction. RT-PCR was performed with the Eppendorf Mastercycler gradient to synthesize cDNA from 50ng/µL of RNA and random primers from the high-capacity cDNA reverse transcription kit (4368814, Applied Biosystems).

A 1:50 dilution of cDNA was combined with EXPRESS SYBR GreenER qPCR Supermix reagents (11784200, Thermofisher Scientific) and 2µM primer pairs of GAPDH, ABCC1 or MAP1B (see Supplementary Table 4 for primer product information). qPCR experiments were run in triplicate with three biological replicates per condition using Applied Biosystems StepOne Plus Real time PCR system. CT values were normalized against corresponding GAPDH CT values using the 2^-ΔCT method and log transformed. A Methods schematic depicts the above described miRNA validation, chemosensitivity, and qPCR assays ( Supplementary Figure 1 ).

All data were analyzed using GraphPad Prism (Version 9.40) software, setting the alpha level at 0.05 for all statistical analyses used. All experiments were completed using biological and technical replicates in triplicate. Two-way repeated measures ANOVAs were conducted to assess group differences across time points or drug doses between cell lines or across transfection conditions. A significant phenotypic change x cell line interaction was followed up with Šídák’s multiple comparison tests. Paired and unpaired t-tests were performed to assess group differences on single dependent measures when appropriate.

Biological sex was not considered due to the limited sample size of females per racial group in the LSCC TCGA dataset. Randomization and blinding of the TCGA cohort were irrelevant because the study was specifically designed to explore differential miRNA expression by race. Power analysis was not performed as this our study is exploratory and data points are limited.

Bioinformatic analysis of the 92 White and 19 Black LSCC patients abstracted from the TCGA revealed 132 out of 1902 miRNAs were significantly differentially expressed (FC >1.5, p < 0.05) ( Supplementary Tables 5 , 6 ) in Black compared with White LSCC. The volcano plot depicts ( Figure 1A ) a slightly greater number of miRNAs that are significantly lower (68 miRNAs) than higher (64 miRNAs) in Black compared with White LSCC patients. Table 1 shows the top 30 miRNAs that are lower and higher in Black compared with White LSCC patients.

We focused on miR-9-5p as it is one of the more abundant miRNAs that also has been characterized as a potential biomarker in head and neck cancer (37). In addition, a survival curve generated by Kmplotter (38) in Figure 1B shows low miR-9 levels correlate with poor overall survival in HNSCC (HR=0.6, logrank p=0.0057). As one of the top 30 lower expressed miRNAs, miR-9 was found to be 4-fold lower (log FC =-1.41) in Black compared with White LSCC patients at p = 0.013.

To explore the influence of miR-9-5p in the context of race, we sought to identify cell lines derived from patients with similar clinicopathologic characteristics, differing only by self-reported race and miR-9-5p levels ( Supplementary Table 1 ). We discovered two cell lines that fit those parameters, UM-SCC-12 (derived from a Black male patient) and UM-SCC-10A (derived from a White male patient) cell lines that were established at the University of Michigan. To assess the expression of mature miR-9-5p in the cell lines, we performed Northern Blot analysis. miR-9-5p was weakly detectable in the UM-SCC-12 cell line. Strong expression of miR-9-5p was detected in the UM-SCC-10A cell line ( Figure 2 ). Endogenous U6, miR-191-5p, miR-16, and Let 7A served as internal loading controls.

To properly interpret our miR-9 validation results, we were required to first determine the intrinsic cellular behavior of the cell lines. We characterized baseline differences in cell migration, cell proliferation and chemosensitivity between our two cell lines.

Cell migration was assessed via the scratch wound assay and operationalized as percentage of wound closure 24h, 48h, and 72h post monolayer disruption. A two-way repeated measures ANOVA was conducted to assess differences in percentage of wound closure across each time point between the two cell lines. UM-SCC-12 cell line had a significantly greater cellular migration compared with UM-SCC-10A cell line at 24h, 48h, and 72h post monolayer disruption (p < 0.0001) by Šídák’s multiple comparisons test ( Figure 3A ).

Cell proliferation was assessed in both cell lines via cell titer blue fluorescence assay at 72h. An unpaired t- test was conducted to assess differences in relative fluorescence between the cell lines. Figure 3B shows that UM-SCC-12 cell line had a greater rate of proliferation at 72 hours compared with UM-SCC-10A cell line (t (4) = 4.96, p < 0.001).

As cisplatin is the primary chemotherapeutic treatment for head and neck cancer, we assessed baseline cisplatin IC50 for both cell lines. Cell line viability was measured at 48h of cisplatin treatment. A two-way repeated measures ANOVA was then conducted to assess differences in cell viability across the cisplatin concentrations between the two cell lines. UM-SCC-12 cell line had decreased sensitivity to cisplatin compared with UM-SCC-10A cell line at the three highest doses of cisplatin treatment: 22.22µM (p < 0.001), 66.67µM (p < 0.0001), and 200µM (p < 0.01) ( Figure 3C ).

An unpaired t-test was conducted on IC50 values generated from three experimental runs between the cell lines. A nonlinear fit of the normalized percentage of cell viability responses relative to the non-transformed cisplatin concentrations was used to calculate the IC50. The calculated IC50 of cisplatin was significantly greater for the UM-SCC-12 cell line at 15.24 um compared with 10.23 um for UM-SCC-10A (t(4) = 5.86, p < 0.01, Figure 3D ).

Collectively, UM-SCC-12 cells had significantly greater baseline cell migration, cell proliferation, and decreased cell killing by cisplatin, and thus a higher cisplatin IC50 when compared to UM-SCC-10A.

We next wanted to understand how increasing miR-9 levels in UM-SCC-12 cell line would alter its cellular phenotype. UM-SCC-12 cells were transiently transfected with a miR-9 mimic or mock oligo control. The scratch wound assay was performed. Figure 4A shows representative images of wound closure captured at 0h, 24h, 48h, and 72h post monolayer disruption for miR-9 mimic and mock transiently transfected cells. By multiple comparisons test, the mimic transfected UM-SCC-12 cells had a significantly lower percentage of wound closure compared to mock transfected control cells at 24h (p < 0.05), 48h (p < 0.01), and 72h post monolayer disruption (p < 0.01) ( Figure 4B ), demonstrating that elevated levels of miR-9 can decrease cellular migration in UM-SCC-12 cell line.

To assess cell proliferation in miR-9 transfected UM-SCC-12 cells, cell titer blue assay was performed on miR-9 transfected UM-SCC-12 cells. A paired samples t-test was conducted on relative fluorescence values across transfection conditions and revealed that the miR-9 transfected UM-SCC-12 cells exhibited lower relative fluorescence compared to mock oligo control, indicating a decrease in cell proliferation (t(2) = 4.65, p < 0.05) ( Figure 4C ).

We proceeded to test the effect of increasing miR-9 levels and chemosensitivity. miR-9 transfected UM-SCC-12 cells were treated with cisplatin and the IC50 was recorded. We noted an increase in cell killing/decreased cell viability in miR-9 transfected cells as compared to mock oligo control at three doses of cisplatin treatment: 2.47µM (p < 0.0001), 7.41µM (p < 0.001), and 22.22µM (p < 0.001) ( Figure 4D ). Furthermore, the calculated IC50 was lower (7.64 µM) compared to than mock oligo control (15.57 µM). Overall, these finding demonstrated that elevated levels of miR-9 can increase UM-SCC-12 sensitivity to cisplatin ( Figure 4E ).

Our next step was to determine if reducing miR-9 levels could produce an opposite phenotype as seen in our miR-9 transfected UM-SCC-12 cells. Here, we knockdown miR-9 levels by treating UM-SCC-10A cells with a miR-9 inhibitor. The scratch assay was performed. Figure 5A shows a panel of representative images of miR-9 inhibited UM-SCC-10A cells at 0h, 24h, 48h, and 72h compared with mock oligo control, showing lower miR-9 levels can significantly increase cell migration at each time point, 24h (p < 0.01), 48h (p < 0.001), and 72h post monolayer disruption (p < 0.001) ( Figure 5B ).

Cell proliferation was evaluated in miR-9 inhibited UM-SCC-10A transfected cells at 72 hours. Relative fluorescence values across transfection conditions was compared via paired t test. Figure 5C shows that cell proliferation was higher in miR-9 inhibited cells relative to the mock oligo control (t(2) = 4.94, p < 0.05).

We next compared chemosensitivity in miR-9 inhibited UM-SCC-10A transfected cells compared to mock oligo control. We found that lowering miR-9 levels decrease cell killing/decreased chemosensitivity at three doses of cisplatin treatment: 2.47µM (p < 0.0001), 7.41µM (p < 0.0001), and 22.22µM (p < 0.05) ( Figure 5D ). The IC50 values was higher (14.57 µM) compared to than mock oligo control (10.71 µM). These findings signify that lower miR-9 levels can decrease cellular sensitivity to cisplatin (t(2) = 28.79, p < 0.01) ( Figure 5E ).

ABCC1 and MAP1B are reported gene targets of miR-9-5p that have been found to predict chemoresistance in cancer (39–42). Thus, as mediators of chemoresistance, we were interested in investigating whether these genes were regulated by miR-9 in LSCC. Initial studies involved determining baseline expression of ABCC1 and MAP1B in UM-SCC-12 and UM-SCC-10A by qPCR. Unpaired t-tests were conducted to assess differences in the log transformed 2^-ΔCT values between the two cell lines. UM-SCC-12 had significantly higher baseline gene expression of ABCC1 (t(4) = 7.69, p < 0.01) and MAP1B (t(4) = 18.69, p < 0.0001) compared with the UM-SCC-10A cell line (t(4) = 7.69, p < 0.01) ( Figures 6A, B ).

Using miR-9 transfected UM-SCC-12 and miR-9 inhibited transfected UM-SCC-10A cells, we sought to determine if miR-9 regulated ABCC1 and MAP1B expression. 24 hours after transfection, gene expression for both ABCC1 and MAP1B was determined by qPCR. Figure 7A shows increased levels of miR-9 in UM-SCC-12 cells significantly decreased ABCC1 gene expression relative to the mock oligo control (t(2) = 4.63, p < 0.05). Conversely, reducing miR-9 levels in UM-SCC-10A significantly increased ABCC1 expression compared with mock oligo control ((t(2) = 7.42, p < 0.05) ( Figure 7B ).

Similar qPCR results were seen with MAP1B whereby increasing miR-9 in UM-SCC-12 cells reduced MAP1B gene expression relative to mock oligo control (t(2) = 6.19, p < 0.05, Figure 8A ). Figure 8B shows that reducing miR-9 in UM-SCC-10A cells resulted in increased levels MAP1B relative to mock oligo control (t(2) = 5.22, p < 0.05). Taken together, these studies suggest that miR-9 regulates ABCC1 and MAP1B levels in LSCC cell lines.