Abstract   Lung  cancer  is  the  leading  cause  of  death  in  the  U.S.  Non-­‐small  cell  lung  cancer  (NSCLC),  the  vast  

majority  of  lung  cancers,  has  three  main  subtypes:  adenocarcinoma,  large  cell  carcinoma  and  squamous  cell  carcinoma.  These  subtypes  are  difficult  to  distinguish  histologically  but  demand  distinct  courses  of  treatment,  and  applying  the  wrong  treatment  can  lead  to  serious  adverse  consequences.  We  have  developed  a  gene  expression-­‐based  classifier  to  distinguish  squamous  and  non-­‐squamous  NSCLC  by  using  quantitative  Nuclease  Protection  Assayi(qNPA).  The  chemistry  does  not  require  RNA  extraction,  reverse  transcription,  or  amplification  and  is  fully  automated  on  the  HTG  Edge  System.      

Results  

Conclusions  We  have  developed  a  robust  classifier  that  can  accurately  distinguish  Squamous  from  non-­‐squamous  NSCLC.  Its  performance  has  been  demonstrated  by  independent  clinical  validation  samples.    

We  first  identified  a  set  of  candidate  biomarker  genes  through  microarray  analysis  of  both  fresh  frozen  and  formalin  fixed  paraffin-­‐embedded  (FFPE)  samples  from  134  NSCLC  patients,  plus  in  silico  analysis  of  six  microarray  datasets  from  GEO.  We  then  refined  the  set  of  marker  genes  and  developed  a  Support  Vector  Machine  (SVM)  classifier  on  an  independent  cohort  of  161  FFPE  samples.  These  161  samples  are  represented  by  variety  of  sources  to  model  the  heterogeneity  of  samples  under  both  academic  and  community  settings.  Finally,  we  validated  the  performance  of  the  classifier  on  another  independent  set  of  97  FFPE  samples.  

Table  1.  Clinical  Data  sets  used  to  develop  and  validate  the  classifier  

Methods  

1877  Non-­‐small  cell  lung  cancer  histological  sub-­‐typing  by  gene  expression  analysis  from  FFPE  tissue  Eva  Wang1,*,  Zhenqiang  Lu1,  Krishna  Maddula1,  Mark  Schwartz1,  Chris  Roberts1,  Mary-­‐Beth  Joshi2,  David  Harpole,  Jr.2,  Vijay  Modur1.    

1HTG  Molecular  Dx,  Tucson,  AZ  85706;  2Duke  University  Medical  Center,  Durham,  NC  *corresponding  author:  ewang@htgmolecular.com  

HTG  Edge  System  &  Chemistry  

Sample Prep Manual

Chemistry

Edge Processor Edge Reader

Automation

The  performance  of  the  classifier  was  measured  by  its  call  concordance  with  three  pathologists’  IHC  panel  consensus  reads.  The  classifier  distinguished  squamous  and  non-­‐squamous  NSCLC  of  the  97  FFPE  independent  samples  with  AUC  of  0.98(Fig  1)  and  accuracy  of  94%(Fig  2).  Two  of  the  six  discordant  samples  were  confirmed  as  positive  in  our  ALK  screening  assay;  ALK  fusions  are  generally  limited  to  adenocarcinomas.  The  classifier,  combining  with  an  ALK  screening  assay  can  provide  increased  accuracy  for  NSCLC  subtyping.  

 The  classifier’s  robustness  was  further  demonstrated  by  diluting  the  tumor  content  with  normal  adjacent  tissues,  with  as  little  as  20%  of  the  original  tumor  content  in  the  final  sample.  All  diluted  samples  were  predicted  correctly(Fig  3).  In  addition,  the  estimated  class  probabilities  did  not  vary  significantly  by  dilution  ratio,  indicating  that  the  classifications  are  robust  to  low  tumor  content.  

 When  adenocarcinoma  and  squamous  lysate  were  titrated  together,  the  prediction  scores  varied  roughly  linearly  with  the  adenocarcinoma  or  squamous  cell  concentration,  reflecting  the  biological  changes  in  the  sample  mixture(Fig  3).    

Table  2.  In  silico  data  sets  used  to  refine  the  classifier  genes.  

Author No. Adeno

No. Large Cell

No. Squamous Total Array Type Data Source

Bild et al 58 53 112 U133 Plus2.0 GEO: GSE3141

Bittner et al 59 34 39 132 U133 Plus2.0 GEO: GSE2109

Hou et al 45 19 27 91 U133 Plus2.0 GEO: GSE19118

Kim et al 62 76 138 U133 Plus2.0 GEO: GSE8894

Su et al 14 14 28 U95A http://www.gnf.org/cancer/epican

Zhu et al 28 10 52 90 U133A GEO: GSE14814

Fig  1  .Final  classifier  performance  on  validation  data  shown  as  ROC.  Area  under  the  curve(AUC)  is  typically  between  0.5  and  1.  

Fig  2.    Final  classifier  performance  on  validation    data.  Samples  on  x-­‐axis  are  sorted  by  prediction  probability.  The  colors  are  based  on  the  final  adjudicated  reads  from  three  pathologists.  

Fig  3.  The  prediction  probabilities  of  mixture  samples:  1)  adenocarcinoma  or  squamous  samples  diluted  with  normal  adjacent  tissue  (NAT)  and  2)  adenocarcinoma  and  squamous  samples  mixed  together  with  different  proportions.  

Data  Sets  

References  i.  Martel  et  al,  Multiplexed  screening  assay  for  mRNA  combining  nuclease  protection  with  luminescent  array  detection.  Assay  and  Drug  Development  Technologies  (2002),1(1),  61-­‐71.  

Presented at AACR Annual Meeting | San Diego, CA | April 2014

*  Samples  from  the  two  discovery  sets  are  Fresh  frozen  and  FFPE  matched  tissues  

Results

  Discovery Set Discovery set Training Set Validation Set Platform Affy qNPA qNPA qNPA

Sample Type Fresh frozen* FFPE* FFPE FFPE Sample size 134 134 161 97

Adenocarcinoma 70 70 74 50 Large cell 22  

Squamous 64  64  65 47

HTG Molecular Diagnostics | 3430 E Global Loop | Tucson, AZ 85706 | (877) 289-2615 | htgmolecular.com