2.1 Case selection

Institutional Review Board approval for this study was obtained from the University of Illinois at Chicago. A case retrieval search was performed at the University of Illinois at Chicago pathology archives for all FNAs and cytology fluid samples that had a corresponding cytology cell block from August 2013 through May 2017. FNA specimens and fluid samples were chosen as all of these cases reflexively have a cell block performed. All samples were initially fixed in CytoLyt and transferred in pellet form into formalin for a minimum of 6 h prior to being placed on the processor in the histology laboratory. A total of 2702 cases were identified during this time period. The diagnoses of these cases were reviewed to identify all cases that had a malignant solid tumor neoplasm (this included carcinomas and soft tissue tumors). A total of 451 cases were identified as being positive for a solid tumor malignancy, and the cell block H&E slides for these cases were reviewed to confirm presence of malignant cells. Approximately 30% of these cases were found to have no malignant cells within the cell block sections and were confirmed to have had the diagnosis of malignancy rendered on the cytospin or FNA smear preparations; these cases were excluded from the study. Another 20% of cases identified were excluded from the study due to cell blocks or slides missing from the archives.

Of the remaining 226 cases, 106 (47%) cases were found to have inadequate numbers of tumor cells on cell block H&E sections (cases with less than 200 tumor cells per H&E section were considered inadequate and not included in the study) or inadequate tissue remaining on the paraffin block for 16 unstained slides to be cut for the molecular analysis performed in this study. A total of 120 cases were identified with enough archival tissue remaining to proceed with molecular analysis (Figure 1).

In addition to the cytology cell blocks for 120 cases, we reviewed the pathology archival material for any surgical pathology case for the patient cohort that had the same diagnosis as the cytology sample. Many of our cases within the cohort were cytologic fluid samples that represented metastatic disease where surgical resection was not an option (Table 1). In addition, our center is a tertiary care referral center and many smaller community practices send patients to our clinics for endoscopic diagnostic procedures, and therefore, resection specimens are not present in our archives for comparison in many lung and gastrointestinal malignancies. Due to these conditions, we were able to find surgical samples for only 18 cases that corresponded to the cytology sample and were included in the study for comparison of NGS results.

A total of 20 negative controls were identified in FNA lymph node samples; 10 negative lymph nodes from patients with no history of malignancy, and 10 negative lymph nodes from patients with a history of malignancy were included.

2.2 DNA and RNA extraction

FFPE cytology cell blocks were sectioned at 7 μm. A total of 15 unstained and unbaked serial sections were used for both DNA and RNA extraction. A 5 μm section was stained with hematoxylin and eosin. Tumor area was marked, and the percentage of tumor cells was estimated by a pathologist. Tumor cells were manually macrodissected to enrich for tumor fraction to achieve ≥20% tumor cells in the DNA/RNA sample. The Promega semi-automated FFPE DNA and RNA extraction kits were used with the Promega Maxwell RSC system (Promega Corporation). Quantitation of DNA and RNA by Qubit (Thermo Fisher Scientific) was performed prior to library preparation.

2.3 Targeted panel sequencing

Library preparation and sequencing using the Oncomine Comprehensive Assay version 3 was performed on all cases, using 20 ng of DNA and 20 ng of RNA for each sample, as has been previously described.⁶ The assay targets 161 unique genes including 84 genes for hotspot mutations, 43 genes for focal copy number gains, 48 full coding sequences for deletion mutations, and 51 fusion drivers. Data analysis was performed using the Torrent Suite software version v5.10. Ion Reporter version v5.10 was used for variant calling. A minimum average depth of coverage of 600X was considered adequate for each sample. Fusion analysis was performed with Ion Reporter version v5.10 fusion analysis workflow. Variants were classified as benign, likely benign, variant of undetermined significance, likely pathogenic or pathogenic based on the clinical criteria set by the College of American Pathologists and the Association of Molecular Pathology.⁷

2.4 Statistical analysis of outcome data

For cytologic and surgical comparison, results from a total of 17 paired samples were included. For each subject, the locus, genes, AA change, and values for variants (surgical and cytology) were recorded. The goal of this analysis was to determine the correlation between variants detected in surgical and cytology samples. Pearson's and Spearman's correlations were measured between variants. Parametric paired t-test was performed to compare the means of variants. Nonparametric paired test (Wilcoxon Signed Rank Test) was used to compare the medians of variants. The significance levels were set at 0.05 for all tests. The SAS 9.4 Version (SAS Institute, Inc.) was used for data management and analyses.

A clinical chart review was performed for all cases that passed quality metrics. A total of 117 subjects were included in this analysis. The charts were reviewed for the following information: date of diagnosis, final pathology diagnosis, treatment regimen, date of recurrence, date of metastasis diagnosis, date of death, and date of last known contact. Time of overall survival (OS) was calculated as the time from study enrollment to death or last contact. Time of progression-free survival (PFS) was calculated as the time from study enrollment to disease progression date, death date, or last contact whichever comes first. The survivor functions for PFS or OS were estimated by Kaplan-Meier survival analysis. Cox proportional hazards model was employed to estimate the adjusted effect of variants, diagnosis, and metastasis status on PFS or OS after adjustment for all other factors.

3.1 Case cohort

A total of 120 cases met the criteria for inclusion in the study. Two samples did not yield enough DNA for library preparation, and one case had too much formalin-induced artifacts for reliable analysis, leaving 117 samples in our study. There were 31 primaries, 6 recurrences, and 80 metastatic cases included in the study (Table 1), and the percentage of tumor cells ranged from 20–100%. The largest portion of the cohort consisted of lung carcinomas (35.8%) including 26 adenocarcinomas, 8 squamous cell carcinomas (SCCs), 6 small cell carcinomas, and 3 poorly differentiated carcinomas, which adequately reflects the frequency of thoracic specimens sent to our university's cytology service. The remaining cases consisted of 6 carcinomas of unknown origin, 9 high-grade serous carcinoma (HGSC) of the ovary and fallopian tube, 1 low-grade serous carcinoma (LGSC), 8 endometrial carcinomas (ECs) (including endometrioid, serous, and clear cell subtypes), 3 cervical carcinoma (1 squamous and 2 adenocarcinoma), 7 breast adenocarcinoma, 9 pancreatic adenocarcinoma, 6 gastrointestinal stromal tumors (GIST), 2 cholangiocarcinoma, 2 gastric adenocarcinoma, 1 low-grade appendiceal mucinous tumor, 7 head and neck SCCs (including larynx, pharynx, ocular, and oral cavity), 2 thyroid carcinoma, 1 salivary gland neoplasms, 2 melanoma, 2 renal cell carcinoma, 1 mesothelioma, 1 urothelial carcinoma, and 3 well-differentiated neuroendocrine tumors. A breakdown of each tumor type is represented in the pie chart (Figure 2).

3.2 Surgical-Cytologic correlation

There were 18 cases within the cohort that had paired surgical and cytology samples. These samples were from different time points over the course of patient treatment (ie: surgical case at primary diagnosis and cytology case at the time of metastasis), and this difference in time was noted in our correlation results table as 1st diagnosis and 2nd diagnosis (Table 2). The Spearman correlation between 1st and 2nd diagnosis was 0.65056 with a p-value of <0.0001, and the Pearson correlation between 1st and 2nd diagnosis was 0.54763 with a p-value of <0.0001. Both methods show a significant correlation between the surgical-cytologic paired samples. A detailed comparison of variants detected in each of the surgical and cytology cases is shown in Figure 3. Ten out of 18 cases had discrepancy in variants between the surgical and cytology samples. In 8 cases, the cytology samples had extra variants compared to surgical samples. All 8 samples were from metastatic disease with additional copy number, missense, or nonsense variants that are likely associated with tumor progression and/or metastasis. In case 15, the surgical sample had a nonsense mutation (R58*) in CDKN2A (VAF 74.9%) and a promoter mutation (c.-146C > T) in TERT (VAF 50.7%) that were not present in the cytology sample. The cytology sample had an indel (W557_K558del) in KIT (VAF 11.3%) and a BRAF V600E (VAF 5.3%) mutation that were not detected in the surgical sample (Figure 3). It is possible that the cytology sample represents a different clone that expanded in the tumor metastasis. Finally, for case 11, there were two surgical samples and one cytology sample. The first surgical sample and cytology samples were both from the primary breast cancer, whereas the second surgical sample was from the metastasis. In this case, the second surgical sample had a pathogenic ESR1 D538G mutation known to be associated with endocrine therapy resistance in breast cancer.⁸ The rest of the variants were identical in both surgical and cytology samples (Figure 3).

3.3 Cytology cases

Within the 117 cases sequenced, the depth of coverage ranged from 648X to 2694X (Figure 4A). A total of 711 variants were detected, including 505 single nucleotide variants (SNVs), 2 multinucleotide variants (MNVs), 73 insertion/deletions (indels), 126 copy number variants (CNVs), and 5 fusions (Figure 4B, Figure 5, Table S1). There was a total of 347 pathogenic variants and 345 variants of undetermined significance (VUS) detected in all cases (Figure 4C). The 20 negative control samples were sequenced with adequate depth of coverage and no pathogenic variants were detected.

In the entire cohort of 117 cases, TP53 was the most frequently mutated (58%) gene (Figure 5). Among the 26 lung adenocarcinomas, 18 cases had clinically actionable variants for which FDA-approved drugs are available. Seven cases had EGFR mutations (4 cases with exon 19 deletion, 1 case with R776H and L858R, 1 case with R831C, and 1 case with P794T). Four cases with KRAS G12C/S/V and TP53 variants, 2 cases with EML4-ALK fusion, 3 cases with BRAF variants (1 V600E, 1 G469V, and 1 exon 15 indel), 1 case with PIK3CA E542K, and 1 case with ERBB2 exon 20 indel. In the remaining 8 cases variants in NF1, STK11, ERBB4, SETD2, BAP1, and CDKN2A genes were detected (Figure 4D, Table S1). One case with KRAS G12S also had a WHSC1L1-FGFR1 fusion. The squamous cell lung carcinomas showed two cases with ARID1A loss of function variants (E1718* and G1848fs), one FGFR3 S249C variant, one MAP2K1 P124L, and one case with PIK3CA amplification and E545A variant. One case of poorly differentiated carcinoma of the lung had an ARID1A G836fs variant. The small cell carcinoma cases and remaining lung carcinoma cases did not have clinically relevant pathogenic variants.⁹

Of the 6 carcinomas of unknown origin, 2 cases showed a KRAS G12D/V variant and one case showed an amplification of the same gene. One case had an ERBB2 amplification with a SMARCA4 loss of function G1232C variant. One case had many amplifications present.

Our cohort of gynecologic tumors included 9 high-grade serous ovarian/fallopian tube cancers (HGSC), one LGSC, 8 ECs, one SCC of the cervix, and two adenocarcinomas of the cervix. Of the HGSC, two cases had a BRCA1 pathogenic variant (W1733* and Q1806*), and one case had a BRCA2 pathogenic variant (L1227fs). Additionally, one case showed a FGFR2 amplification and another case showed a FGFR1 pathogenic variant (D166del). The LGSC showed a PIK3CA-FNDC3B fusion and a KRAS G12V variant. Of the 8 ECs, two had the most common hotspot PIK3CA variants (Q546K and E542A), one case with two different hotspot PIK3CA variants (E81K and R88Q) and an additional case with a PIK3CA amplification. One case of EC had a BRCA1 frameshift variant (E1210fs), and one case showed ERBB2 amplification. FGFR3 amplification and gain-of-function variant in one case appeared to be the driving alterations. Within the 7 breast cancer cases, two cases showed a BRCA2 nonsense variant (p.R2318* and p.K3326*), one case showed ERBB3 G234R, one case showed KRAS G12V, and one case showed ARID1A P453fs. No PIK3CA, BRCA1, or ERBB2 alterations were identified in the breast carcinoma cases.^{10, 11} Amplifications of AKT1, MYCN, MYC, FGFR1, FGFR4, FGF3, FGF19, IGF1R, CCND1, CCNE1, TERT, and NTRK3 were detected in 4/7 cases (Figure 5).

All nine pancreatic adenocarcinomas harbored exon 2 pathogenic gain-of-function variants in KRAS and 6/9 cases had a deleterious loss-of-function variant in TP53. In addition to these variants, one case showed a frameshift mutation in BRCA1 (Q867fs), and another case showed a less common G1049R gain-of-function variant in PIK3CA. All six cases of GIST harbored a pathogenic variant in KIT—three indels including a splice site variant and 3 gain of function missense variants. Of the two cholangiocarcinoma cases, one case had a SMAD4 R361H pathogenic variant and the other case had a pathogenic KRAS, TP53, and additional CTNNB1 S45F variant. The two gastric carcinoma cases showed an ERBB2 R678Q pathogenic variant with a PIK3CA E542K and KRAS G12V variants.

Within the SCCs of the head and neck region, there was variability in locations that was reflected in the variants detected. One case from the face showed two loss of function variants in BRCA1 (E17758* and E1491*). A frontal sinus-based neoplasm had a loss of function BRCA2 K3326* variant. A nasopharyngeal-based lesion harbored a pathogenic HRAS Q61R variant and a FGFR1 amplification. A laryngeal SCC was found to have a PIK3CA H1047R pathogenic variant. The additional larynx-based tumor and the floor of mouth tumor included in our cohort showed significant number of copy number variations (CNVs).¹²

In the smaller diagnosis cohort, one melanoma case had a NRAS Q61R variant and the other case of melanoma had a BRAF V600E variant. Both papillary thyroid carcinoma cases harbored BRAF V600E variants. One case of renal cell carcinoma harbored a KRAS G12D variant and showed an oncocytic papillary phenotype. The salivary gland malignancy showed PIK3CA H1047R and HRAS Q61R pathogenic variants.

Of note, there were no significant pathogenic variants detected in the urothelial carcinoma, the mesothelioma, the low-grade appendiceal mucinous tumor, and neuroendocrine tumors. A summary of variants along with diagnosis and clinical relevance is shown in Figure 5, and for a full list of variants detected in our cohort, please refer to Table S1.