Article Text

Download PDFPDF

Pre-treatment patient selection for nivolumab benefit based on serum mass spectra
  1. Jeffrey Weber1,
  2. Alberto J Martinez1,
  3. Heinrich Roder2,
  4. Joanna Roder2,
  5. Krista Meyer2,
  6. Senait Asmellash2,
  7. Julia Grigorieva2,
  8. Maxim Tsypin2,
  9. Carlos Oliveira2,
  10. Arni Steingrimsson2,
  11. Kevin Sayers2,
  12. Antonella Bacchiocchi3,
  13. Mario Sznol4,
  14. Ruth Halaban3 and
  15. Harriet Kluger4
  1. Aff1 grid.170693.a000000012353285XH. Lee Moffitt Cancer Center Tampa FL USA
  2. Aff2 Biodesix Inc. Boulder CO USA
  3. Aff3 grid.47100.320000000419368710Yale University School of Medicine New Haven CT USA
  4. Aff4 grid.433818.5Department of Medical OncologyYale Cancer Center New Haven CT USA

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Meeting abstracts


The durability of anti-tumor responses observed in patients treated with antibodies blocking PD-1 has provided a central role for these drugs in melanoma therapeutics. Identifying predictive biomarkers to aid therapeutic decision making is critical for realizing the full potential of these immunotherapies. We report on the development of a pre-treatment serum test to separate melanoma patients into two groups with significantly different outcomes following nivolumab therapy.


Pre-treatment serum samples were available from 119 patients in the NCT01176461 study (“SET1”) and 30 patients from an observational study (“SET2”) at two institutions. All patients had advanced un-resectable melanoma and received nivolumab.

Mass spectra were collected from all samples using the “deep MALDI” approach [1]. We identified 351 mass spectral peaks for use in classifier construction. SET1 was split into a development (DEV) (N=60) and an internal validation (VAL) (N=59) set. Deep learning methods were used to construct a classifier correlating with time-to-event data in a fashion similar to Roder et al [2] using only the DEV set. This classifier was validated on the VAL set and a test was constructed using the same procedure with the whole SET1 and performance evaluated on the independent SET2.


The test separated the populations into two groups, “Early”/”Late”, with worse/better outcome on nivolumab treatment. The hazard ratios (HRs) between Early and Late groups are presented in Table 1. Test classification groups did not show any association with available PD-L1 expression data and remained significant in multivariate analysis.

Table 1


We have constructed a test to identify melanoma patients most likely to have improved survival on nivolumab therapy. The test validated in an independent sample set with HR~0.3 and appears to be independent of PD-L1 expression. Some proteins used in the test are related to acute Phase reactions and the complement system. While further validation and protein identification studies are needed, this test may become a clinically useful predictive biomarker for nivolumab therapy.

This research was supported in part by the Yale SPORE in Skin Cancer, funded by the NCI, NIH, under award number 1 P50 CA121974 (R.H.)


  1. 1.
  2. 2.